JP2017093626A - Self-travelling electronic apparatus - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an electronic apparatus that collates an image that the self-travelling electronic apparatus takes at a target position with a position information database for storing images taken at a plurality of positions in a travelling region, and can autonomously travel to the target position.SOLUTION: A self-travelling electronic apparatus includes a travelling control part 11a, a position detection part 11b for detecting a travelling position, an imaging part 133 for imaging the surrounding, a feature point extraction part 11c for extracting a plurality of features from the image, a position information database storage part 61a for storing position information registered by associating the feature points possessed by the imaging part with the imaging positions beforehand, a query storage part 61b for storing the feature point at a target position, and a target position determination part 11d for collating an arrangement relation between the feature point stored in the query storage part and the feature point included in each piece of the position information stored in the position information database storage part, and determining the target position based on the position information that includes the largest number of feature points in the corresponding arrangement relation. The travelling control part controls the travelling to the target position based on the target position.SELECTED DRAWING: Figure 1

Description

  The present invention relates to a self-propelled electronic device, and more specifically, creates an image database by photographing surrounding scenes at a plurality of positions in a traveling region, and photographs the surrounding scenes at a target position to be reached. The present invention relates to a self-propelled electronic device that travels to a target position by collating with the image.

2. Description of the Related Art Self-propelled electronic devices are known that are equipped with a sensor that detects surrounding obstacles and the like and that autonomously travel and work. For example, there are vacuum cleaners, air cleaners, and security devices. As one of the sensors or for other purposes, a self-propelled electronic device having a camera that captures a surrounding scene is known.
The self-propelled electronic device has various modes regarding traveling in order to control traveling according to the purpose of the work area. As an example, a self-propelled cleaner has an “automatic operation mode”, a “wall-side mode”, and a “spot mode” with respect to control of traveling. The “automatic operation mode” is a mode for cleaning the entire house when absent. The “wall-side mode” is a mode in which the self-propelled cleaner performs cleaning while moving along the wall. By selecting “Border mode”, you can clean the wall. The “spot mode” is a mode for cleaning only the periphery of the area designated by the user. A simple way to specify an area is for the user to carry a self-propelled cleaner to the area.

  Several techniques related to the traveling control according to the situation as described above have been proposed. For example, in order to select and clean the area to be cleaned from a plurality of cleaning areas, the current position is identified from the output of the autonomous driving sensor and the stored map information, and autonomously up to the cleaning area selected based on the current position. A traveling robot cleaner has been proposed (see, for example, Patent Document 1). In addition, it communicates with a server device that searches for a travel route by specifying a region with the fewest number of times of travel, and preferentially travels in a region that has been traveled in the past among the regions specified by the user. A self-propelled electronic device to be calculated has been proposed (see, for example, Patent Document 2).

JP 2002-85305 A JP 2014-85829 A

In the “automatic operation mode” described above, when the user gives an instruction with a single touch, for example, by pressing the cleaning start button on the remote control, the self-propelled electronic device detects the size of the room and the location of the obstacle in response to the instruction. You can drive. In the “wall-side mode” as well, when the user gives an instruction with a single touch, the self-propelled electronic device can detect the position of the wall and run in the area along the wall.
However, in the “spot mode”, the self-propelled electronic device cannot automatically guess the area intended by the user. As a simple method of instructing the intended area to the self-propelled electronic device, the user moves the self-propelled electronic device to the position where the work such as cleaning is desired, and then instructs the start of the work in the spot mode by using a remote control or the like. Given this, the region can be specified. As described above, in the spot mode, it is necessary to move the self-propelled device to a place where the work is desired before the work is performed. It can be said that it is inferior to other modes in terms of convenience.

There is also a method in which a user designates a position using a mobile communication device such as a smartphone as described in Patent Documents 1 and 2 instead of carrying and moving a self-propelled electronic device.
For example, a self-propelled cleaner has a gyro sensor that detects the direction of movement, an encoder that detects the amount of movement, an autonomous traveling unit that detects obstacles and autonomously travels the self-propelled electronic device, and a smartphone that the user has It is assumed that communication means such as a wireless LAN for communication is provided. When the relative position from the charging base, which is the base point, is instructed to the self-propelled electronic device configured in this way from the user's smartphone, control is performed in which the self-propelled electronic device moves to the target position in response to the instruction. Conceivable.
Here, the smartphone that gives instructions to the self-propelled electronic device has an application for displaying a map showing the layout of the room and the arrangement of “things” such as furniture and home appliances as obstacles on the screen. It is possible.

  Allows the app to indicate the relative distance from the home position when the user touches the location on the map that the self-propelled cleaner wants to clean. In such a system, when the screen is touched, the relative position between the charging stand and the position that the self-propelled cleaner wants to clean is calculated from the positional relationship between the touched position on the screen and the charging stand on the screen. And transmit to the self-propelled electronic device. On the self-propelled electronic device side, the autonomous traveling unit causes the self-propelled electronic device to travel to the location indicated by the relative position based on the relative position received from the smartphone.

However, there is one problem here. In such a system, it takes time and effort for the user to create a floor plan and a map for arranging objects. Furthermore, if the position of the thing shown on the map is not in the same positional relationship as the actual thing, the map cannot be designated with high accuracy. That is, the position information on the map must be correctly associated with the actual arrangement of objects in the room. Therefore, when creating a map, it is necessary to measure the positional relationship and the distance interval between objects in an actual room. To make it easy for anyone to use, it is desirable to simplify the process of creating map information.
Furthermore, in such a system, the position cannot be instructed to the self-propelled cleaner unless the user has a smartphone. Users who do not have a smartphone cannot use this system. Some users are not good at operating smartphones. For such users, this system is difficult to use, and improvement is also desired from the viewpoint of convenience.

On the other hand, considering normal human behavior, an instruction such as “clean the front of the sofa” is common among humans. It is desirable that such an intuitive instruction can be given to an electronic device.
In recent years, speech recognition technology has advanced, and it has become possible to accurately recognize the intention of a speaker. That is, the intention of the instruction “clean the front of the sofa” can be easily identified by a machine. Therefore, if the self-propelled electronic device recognizes where “in front of the sofa” is and sets the position as a target position and can autonomously travel, an intuitive and simple instruction is possible.

The present invention has been made in view of the above circumstances, and is a position information database that stores images taken by a self-propelled electronic device at a target position and images taken at a plurality of positions in a running area. And providing an electronic device that can autonomously travel to the target position.
Furthermore, if the image captured at the target position is associated with an intuitive instruction such as voice, an intuitive instruction can be given to the self-propelled electronic device regarding the position. For example, position information that can be intuitively understood by humans, such as “in front of the sofa”, is important for recognizing in association with position information that can be recognized by the self-propelled electronic device, that is, relative position information from the base point. It provides the elemental technology.

  The present invention relates to a travel control unit that controls travel, a position detection unit that detects a position that changes with the travel, an imaging unit that captures a surrounding scene, and a plurality of feature points extracted from the captured image. Position information that stores in advance the position information registered in association with the feature points relating to the images captured by the image capturing unit at a plurality of positions in the traveling area and the positions where the image capturing is performed A database storage unit, a query storage unit that stores feature points related to an image captured by the imaging unit at a target position, a feature point stored in the query storage unit, and a position information database storage unit A target position determining unit that collates the positional relationship with the feature points included in each positional information and determines a target position based on positional information that includes the most feature points in the corresponding positional relationship, Row control unit provides a self-propelled electronic device for controlling the travel to its target position on the basis of the target position and that the position detecting unit is determined as the current position to be detected.

  In the self-propelled electronic device according to the present invention, the query storage unit stores feature points related to an image captured by the imaging unit at a target position, and the target position determination unit is stored in the query storage unit. The feature point and the feature point at each position stored in the position information database storage unit are collated, the position information associated with the corresponding feature point is set as target position information, and the travel control unit is the position detection unit. The travel to the target position is controlled based on the current position detected by the camera and the determined target position, so that the self-propelled electronic device captures an image captured at the target position and a plurality of positions within the travel area. It is possible to autonomously travel to the target position by collating with a position information database storing stored images.

It is a block diagram which shows the electrical structure of the self-propelled cleaner which concerns on embodiment of this invention. It is an external appearance perspective view of the self-propelled cleaner which is one embodiment of this invention. It is a bottom view of the self-propelled cleaner shown in FIG. It is a vertical sectional view along the front-back direction of the self-propelled cleaner shown in FIG. It is a horizontal sectional view of the self-propelled cleaner shown in FIG. It is a top view which shows the position of the home position where the self-propelled cleaner concerning this embodiment is docked to the charging stand. It is a top view which shows a mode that the self-propelled cleaner shown to FIG. 6A exists in the position away from the home position. In this embodiment, it is explanatory drawing which shows an example and division of the floor plan in which a self-propelled cleaner runs. It is explanatory drawing which shows a mode that the self-propelled cleaner in this embodiment starts from a home position and sequentially performs photographing in each section when creating a position information database. In an embodiment, it is a flow chart which shows the whole processing performed when a device control part creates a position information database. In an embodiment, it is a flow chart which shows creation processing of feature point information in apparatus control part in position information database creation. In an embodiment, a device control part is a flow chart which shows creation processing of distance information in position information database creation. It is explanatory drawing which shows the data structure of the positional information database produced in this embodiment. It is explanatory drawing which shows the pixel pattern detected as a feature point in this embodiment. In this embodiment, it is explanatory drawing which shows visually the position where a feature point is extracted from the image where the table as an example of an object was image | photographed. In this embodiment, it is a flowchart which shows the whole process of the estimation of the position which an apparatus control part performs. It is a flowchart which shows the procedure of the matching process shown in FIG. It is a flowchart which shows the detail of the determination process of the reference point shown in FIG. It is a flowchart which shows the search process of the reference point pair shown in FIG. 17, ie, the process which searches a corresponding reference point. It is a flowchart which shows the search process of the corresponding point shown in FIG. In this embodiment, it is the 1st explanatory view showing visually the decision processing of the reference point used for the feature point matching. In this embodiment, it is the 2nd explanatory view which shows visually the decision processing of the reference point used for the feature point matching. In this embodiment, it is explanatory drawing which shows visually the image search by the matching of a feature point. In this embodiment, it is explanatory drawing which shows the concept of the score to calculate visually. In this embodiment, it is explanatory drawing which shows arrangement | positioning and imaging | photography position of the image | photographed room. In this embodiment, it is explanatory drawing which shows an example of the image image | photographed in the candidate position in the room shown in FIG. It is explanatory drawing which shows the example of the image image | photographed at the position (1) shown in FIG. It is explanatory drawing which shows the image image | photographed in the same position as FIG. 24A. It is explanatory drawing which shows the example of the image image | photographed at the position (2) shown in FIG. It is explanatory drawing which shows the image image | photographed in the same position as FIG. 24A. It is explanatory drawing which shows the example of the image image | photographed at the position (3) shown in FIG. It is explanatory drawing which shows the image image | photographed in the same position as FIG. 24A. It is explanatory drawing which shows the example of the image image | photographed at the position (4) shown in FIG. It is explanatory drawing which shows the image image | photographed in the same position as FIG. 24A. It is explanatory drawing which shows the example of the image image | photographed at the position (5) shown in FIG. It is explanatory drawing which shows visually a mode that the feature point shown to FIG. 24A is extracted from an image.

Hereinafter, the present invention will be described in more detail with reference to the drawings. In addition, the following description is an illustration in all the points, Comprising: It should not be interpreted as limiting this invention.
(Embodiment 1)
≪Specific modes of self-propelled electronic devices≫
First, a self-propelled cleaner will be described as an example of the self-propelled electronic device of the present invention.
FIG. 1 is a block diagram showing an electrical configuration of a self-propelled cleaner according to an embodiment of the present invention. 2 is an external perspective view of the self-propelled cleaner according to the embodiment of the present invention, and FIG. 3 is a bottom view of the self-propelled cleaner shown in FIG. FIG. 4 is a vertical sectional view along the front-rear direction of the self-propelled cleaner shown in FIG. FIG. 5 is a horizontal sectional view of the self-propelled cleaner shown in FIG. 2.

≪Configuration of self-propelled vacuum cleaner≫
As shown in FIGS. 2 and 3, the self-propelled cleaner 1 according to the first embodiment includes a flat plate-shaped housing 2. In the case of the first embodiment, the housing 2 has a disk shape, but is not limited to this, and the shape in plan view may be an elliptical shape or a polygonal shape such as a quadrilateral, a pentagon, or a hexagon.
The housing | casing 2 contains the top plate formed in circular. The top plate is composed of a top plate front portion 2b ₁ constituting the front portion and a lid portion 2b ₂ constituting the middle portion to the rear portion, and the lid portion 2b ₂ is on the side of the boundary with the top plate front portion 2b ₁ The hinge part (not shown) arranged in the part is opened upward using a fulcrum as a fulcrum. A plurality of air holes 2b ₁₁ for releasing heat of the circuit board 11S disposed inside are formed at the front end of the top front part 2b ₁ .

Moreover, the housing | casing 2 contains the side plate formed in cyclic | annular form. As shown in FIGS. 4 and 5, the housing 2 includes an internal structure wall 2d. Each of the side plates includes an arc-shaped side plate front half 2c ₁ and a side plate rear half 2c ₂ . The side plate front half 2c ₁ is movably fitted to the internal structural wall 2d via a resilient member (not shown) so as to function as a bumper. The side plate front half 2c ₁ is provided with a collision sensor 14C for detecting a collision of the side plate front half 2c ₁ (not shown in FIGS. 2 to 5). Further, in the side plate front half 2c ₁ , the ultrasonic receivers 14A are arranged at three positions on the front and diagonally forward sides, and the ultrasonic transmitters 14B are provided at two positions between the three ultrasonic receivers 14A. Has been placed.
Further, an induction signal receiving unit 24, an imaging unit 133, and a charging connection unit 13 are provided at positions that can be seen from the outside of the front surface of the housing 2. Charging connection portion 13 includes a connection terminal to charge base 201 described later and a wiring connecting the connection terminal and rechargeable battery 12.

The housing 2 has a suction port 31 provided at the bottom, a first exhaust port 32 provided obliquely above the rear portion, and a second exhaust port 33 provided at the outer peripheral portion. A dust collector 15 and an electric blower 115 are arranged. The dust collection unit 15 is a part that collects indoor dust, and includes a dust collection container 15a and a dust collection filter 15b. The dust collection container 15 a is formed with an inflow port that leads to an inflow passage that communicates with the suction port 31 and an exhaust port that communicates with a duct portion 114 that communicates with the electric blower 115.
The 2nd exhaust port 33 is opened toward the front, and is comprised so that the airflow which passed the dust collection part 15 and the electric blower 115 may be blown out toward the front. Here, “front” means a direction parallel to the direction in which the self-propelled cleaner goes straight, but it does not have to be strictly a parallel direction, and includes some oblique directions.

In the front half of the bottom surface of the self-propelled cleaner 1, there are a rotating brush 9 disposed in the back of the suction port 31, a side brush 10 formed diagonally forward and leftward of the suction port 31, and an obliquely rearward side of the suction port 31. A left drive wheel 22L and a right drive wheel 22R, which will be described later, formed at the positions are provided. The rotating brush 9 and the side brush 10 are driven and rotated by a brush motor 119. A rotatable rear wheel 26 is provided at an intermediate position in the left-right direction of the rear half of the bottom surface. The wheels of the rear wheel 26 are rotatable. In FIGS. 3 and 4, a state where the rear wheel 26 is rotated 180 ° forward is indicated by a two-dot chain line.
While the self-propelled cleaner 1 is self-propelled on the floor surface where it is installed, the self-propelled cleaner 1 sucks in air containing dust on the floor surface (travel surface) and exhausts the air from which the dust has been removed. clean. The self-propelled cleaner 1 has a function of autonomously avoiding obstacles detected by the obstacle detection unit 14 and autonomously returning to a charging station (not shown) when cleaning is completed.
In this specification, the floor is not limited to a wooden flat floor. Even if it is a carpet, a mat, a tatami mat, etc., it should just be a surface which self-propelled cleaner 1 can run. The floor surface is not limited to a flat surface, and may be a surface that is curved to some extent, an uneven surface, or a stepped surface as long as it can travel.

As shown in FIG. 1, the self-propelled cleaner 1 includes a drive unit 23, a circuit board 11 </ b> S, a failure detection unit 14, and a rechargeable battery 12.
The drive unit 23 includes a left wheel drive motor 21L, a right wheel drive motor 21R, a left drive wheel 22L, and a right drive wheel 22R. The self-propelled cleaner 1 has a left wheel drive motor 21L, a right wheel drive motor 21R, a left wheel encoder 71L, a right wheel encoder 71R, a left drive wheel 22L, and a right drive wheel 22R. If the left wheel drive motor 21L and the right wheel drive motor 21R are stepping motors, the number of motor drive pulses may be used instead of the encoder. In this embodiment, description will be made assuming that the left wheel encoder 71L and the right wheel encoder 71R actually include encoders.

The circuit board 11S includes a device control unit 11, a storage unit 61, a left wheel driver 121L, a right wheel driver 121R, a gyro sensor 20, a blower driver 123, and a brush driver 125.
The left wheel driver 121L electrically drives the left wheel drive motor 21L, and the right wheel driver 121R electrically drives the right wheel drive motor 21R. The gyro sensor 20 detects a change in the traveling direction. The blower driver 123 electrically drives the electric blower 115. The brush driver 125 electrically drives the brush motor 119.

The device control unit 11 controls the left wheel drive motor 21L, the right wheel drive motor 21R, the ion generator 120, the electric blower 115, and the brush motor 119 via a driver.
Moreover, the apparatus control part 11 includes the function of the traveling control part 11a which controls driving | running | working of the self-propelled cleaner 1, the position detection part 11b, the feature point extraction part 11c, and the target position determination part 11d.
The travel controller 11a controls the travel of the self-propelled cleaner 1 by controlling the rotation of the left wheel drive motor 21L and the right wheel drive motor 21R using the encoder signals from the left wheel encoder 71L and the right wheel encoder 71R. .
The position detection unit 11b acquires the distance and direction that the self-propelled cleaner 1 has traveled from the home position based on the encoder signal. In this embodiment, the home position is a position in the room of the self-propelled cleaner 1 in a state where the self-propelled cleaner 1 is docked on the charging stand 201 described later, and is a position detected by the position detection unit 11b. This is the reference position.

FIG. 6A is a plan view showing the position of the home position where the self-propelled cleaner 1 according to this embodiment is docked on the charging stand 201. FIG. 6B is a plan view showing a state in which the self-propelled cleaner 1 shown in FIG. 6A is located away from the home position. In FIG. 6A and FIG. 6B, the center position when the self-propelled cleaner 1 is at the home position is indicated by the origin “O” (O) of the room coordinates. Further, an x-axis of the coordinate axis extending in the horizontal direction from the room coordinate origin and a y-axis of the coordinate axis extending in the vertical direction are shown. 6A and 6B, the self-propelled cleaner 1 and the charging stand 201 exist, whereas the origin O, the x-axis, and the y-axis are conceptual.
FIG. 6B shows that the self-propelled cleaner 1 is at a position of minus 15 centimeters in the x-axis direction and 20 centimeters in the y-axis direction from the room coordinate origin.
The position detector 11b detects the position of the self-propelled cleaner 1 with the coordinate values in the x and y directions with reference to the home position shown in FIG. 6A.
The device control unit 11 is a part that controls the operation of each component of the self-propelled cleaner 1, and is mainly realized by a microcomputer including a CPU, a RAM, an I / O controller, a timer, and the like.

The CPU executes processing based on a control program stored in advance in a storage unit 61, which will be described later, and developed in the RAM, and organically operates each hardware to execute the cleaning function, the traveling function, and the like of the present invention.
The storage unit 61 is a memory that stores information necessary for realizing various functions of the self-propelled cleaner 1 and a control program. In this embodiment, a flash memory and a RAM are used in combination as the nonvolatile memory. The storage unit 61 includes a location information database storage unit 61a and a query storage unit 61b.
Although not illustrated in FIG. 1, the storage unit 61 includes selection instructions used when selecting feature points related to an image captured at candidate positions that are candidates for target positions, with each candidate position as a target. You may include the candidate position memory | storage part which each associates and memorize | stores.

The failure detection unit 14 includes an ultrasonic reception unit 14A, an ultrasonic transmission unit 14B, a collision sensor 14C, and a floor surface detection sensor 18. The floor surface detection sensor 18 is disposed at the center in the left-right direction at the bottom of the housing 2 and at the front end and the axial center position of the left and right side brushes 10.
The obstacle detector 14, particularly the ultrasonic receiver 14A and the ultrasonic transmitter 14B, detects that the self-propelled cleaner 1 is in contact with or approaching an obstacle such as an indoor wall, desk, or chair while traveling. Part. It is used for the self-propelled cleaner 1 to travel along the wall or obstacle while detecting it. Further, the ultrasonic receiving unit 14A can measure the approximate distance to the obstacle. The obstacle detector 14 detects proximity to the obstacle using the ultrasonic receiver 14A and the ultrasonic transmitter 14B. Instead of the ultrasonic receiving unit 14A and the ultrasonic transmitting unit 14B, or together with the ultrasonic receiving unit 14A and the ultrasonic transmitting unit 14B, other types of non-contact sensors such as an infrared distance measuring sensor may be used.
The collision sensor 14 </ b> C is disposed, for example, inside the side plate front half 2 c ₁ of the housing 2 in order to detect that the self-propelled cleaner 1 has come into contact with an obstacle during traveling. The CPU knows that the side plate front half 2c ₁ has collided with the obstacle based on the output signal from the collision sensor 14C.

Each floor surface detection sensor 18 detects a large level difference such as a descending staircase that the self-propelled cleaner 1 should avoid and travel.
The CPU recognizes a position where an obstacle or a step exists based on the signal output from the obstacle detection unit 14. Based on the position of the recognized obstacle or step, the direction to travel next is determined while avoiding the obstacle or step. The left and right floor surface detection sensors 18 detect the descending stairs when the front floor surface detection sensor 18 fails to detect the level difference, for example, when the floor level detection sensor 18 has entered the diagonal direction with respect to the level difference, and the self-propelled cleaner 1 Prevent falling to the downstairs.
The gyro sensor 20 provides travel direction information to the travel control unit 11a when the self-propelled cleaner 1 travels.
Furthermore, the self-propelled cleaner 1 includes an induction signal receiving unit 24, a charging connection unit 13, a notification unit 55, an electric blower 115, a brush motor 119, and an ion generator 120. Furthermore, a wireless communication unit 131, an imaging unit 133, and a voice recognition unit 135 are provided.
The left wheel drive motor 21L rotates and stops the left drive wheel 22L, and the right wheel drive motor 21R rotates and stops the right drive wheel 22R. By configuring the drive motor so that the left drive wheel 22L and the right drive wheel 22R can be independently rotated in both forward and reverse directions, the traveling state of the self-propelled cleaner 1 such as forward, backward, turning, acceleration / deceleration, etc. Realized.

The traveling control unit 11a causes the right driving wheel 22R and the left driving wheel 22L to rotate forward in the same direction, and causes the self-propelled cleaner 1 to travel forward in which the central ultrasonic receiving unit 14A is disposed. Further, the left driving wheel 22L and the right driving wheel 22R are rotated backward in the same direction and moved backward, and rotated in the opposite directions or rotated at different speeds to turn the self-propelled cleaner 1. The different speeds include the case where one drive wheel stops.
For example, when the traveling control unit 11a determines that each sensor of the failure detection unit 14 has reached the periphery of the cleaning area, the traveling control unit 11a decelerates the left driving wheel 22L and the right driving wheel 22R and then stops them. After that, the left driving wheel 22L and the right driving wheel 22R are rotated in opposite directions to turn the self-propelled cleaner 1 by 90 °, advance by a distance equal to the opening width of the suction port 31, and further turn by 90 °. Drive in the opposite direction to the course of The self-propelled cleaner 1 is zigzag-running by repeating the above-mentioned turning in a zigzag shape, and is controlled so that the entire cleaning area can be cleaned.
Furthermore, since the self-propelled cleaner 1 includes the left wheel encoder 71L and the right wheel encoder 71R, the self-propelled cleaner 1 travels by detecting how much the left driving wheel 22L and the right driving wheel 22R have rotated. It is possible to detect the turning angle and the distance moved. That is, given the distance and direction from the home position to the target position, it is possible to move autonomously to the target position.

Further, when the obstacle detection unit 14 detects an obstacle on the course, the traveling control unit 11a turns the self-propelled cleaner 1 after decelerating or stopping and turns to avoid the obstacle. When the obstacle is detected until it is no longer detected, the self-propelled cleaner 1 is turned in a direction to approach the extension of the original course, and the traveling is continued. When each floor surface detection sensor 18 no longer detects the floor surface, the self-propelled cleaner 1 is temporarily stopped and moved backward and / or turned to control traveling so as not to fall from a step such as stairs. In this way, the travel control unit 11a causes the self-propelled cleaner 1 to travel while avoiding obstacles over the entire installation location or the entire desired range.
In this embodiment, the front means the forward direction of the self-propelled cleaner 1, that is, the direction from the rear wheel 26 to the suction port 31 in FIG. 3, and the rear means the self-propelled cleaner 1. 3, that is, the direction from the suction port 31 toward the rear wheel 26 in FIG. 3.

The induction signal receiving unit 24 is an infrared sensor for receiving infrared rays, and is disposed in the front part of the housing 2. The induction signal receiving unit 24 receives a position indicator signal (beacon) and the like emitted from the induction signal transmission unit 203 of the charging stand 201 as a docking station. Furthermore, you may receive the infrared signal from the remote control apparatus which is not illustrated in FIG. The device control unit 11 recognizes the received infrared signal from the remote controller and responds to an instruction given by the user.
The device control unit 11 detects a signal emitted from the induction signal sending unit 203 of the charging stand 201 by the induction signal receiving unit 24 and recognizes the direction in which the charging stand 201 is located. The travel control unit 11a travels in the direction in which the charging stand 201 is located when cleaning is completed, when the remaining charge of the rechargeable battery 12 is reduced, or when a predetermined cleaning operation period has elapsed. The self-propelled cleaner 1 is returned to the charging stand 201. However, if there is an obstacle, it is moved in the direction of the charging base 201 while avoiding it.
The notification unit 55 notifies the user of the state of the self-propelled cleaner 1. In this embodiment, the notification unit 55 is a sound output circuit that outputs sound and a speaker. In another embodiment, the upper portion of the housing 2, may be more specifically, provided with a display unit using an LED lamp or the like, for example, top plate front portion 2b ₁ and the side plate half portion 2c _2. As yet another aspect, an information processing apparatus having a communication function such as a smartphone owned by the user may be notified through the wireless communication unit 131.

The wireless communication unit 131 is a wireless communication circuit in which the self-propelled cleaner 1 communicates with an external device. A typical example is a wireless LAN communication circuit. Alternatively or in addition, communication may be performed using a mobile communication method such as LTE. The external device is, for example, a smartphone used by the user. The wireless communication unit 131 receives a wireless signal related to an instruction given by a user using the smartphone. The device control unit 11 recognizes the received radio signal and responds to the instruction.
Although not illustrated in FIG. 1, the wireless communication unit 131 may include a function as an instruction receiving unit that receives a travel instruction to travel to a target.
A specific mode of the imaging unit 133 is a device built-in camera module. It may be possible to shoot not only still images but also moving images. As shown in FIG. 2, the imaging unit 133 is attached above the induction signal receiving unit 24 and photographs the scenes above and around the self-propelled cleaner 1. The camera module as the imaging unit 133 includes a wide-angle lens, and photographs a side view as well as the upper side.
The voice recognition unit 135 is a microphone and a voice recognition circuit that recognizes whether or not a predetermined instruction is detected by detecting a voice uttered by the user. Although not shown in FIG. 1, the voice recognition unit may include a function as an instruction receiving unit that receives a travel instruction to travel to a target. That is, the function as the instruction receiving unit may be realized by the wireless communication unit 131 described above, or may be realized by the voice recognition unit 135, and both may realize the function of the instruction receiving unit.
The above is description of the structure of the self-propelled cleaner concerning this embodiment.

A few things about the charging stand. The charging stand 201 includes a charging terminal unit 202 and an induction signal sending unit 203. The self-propelled cleaner 1 is docked on the charging stand 201, that is, the charging terminal portion 202 of the charging stand 201 and the charging connection portion 13 of the self-propelled cleaner 1 are in electrical contact with each other. The vacuum cleaner 1 is supplied with electric power from the charging stand 201. The rechargeable battery 12 of the self-propelled cleaner 1 is charged while being docked on the charging stand 201. The induction signal transmission unit 203 includes a signal generation circuit that generates a beacon signal and an LED that emits the generated infrared light signal.
In general, the charging stand 201 is often placed in an unobstructed place such as a corner of a room, and once the placement position is determined, it is often placed at that position. When cleaning is completed, the self-propelled cleaner 1 returns to the charging stand 201 for charging and charges the rechargeable battery 12. That is, the self-propelled cleaner 1 stands by while charging the rechargeable battery 12 at the charging stand when not cleaning.

≪Overview of location information database creation≫
Next, an outline of a procedure for creating a position information database in this embodiment will be described.
The position information database is created by using, as a database, surrounding scenes photographed by the imaging unit 133 at a plurality of positions in the room where the self-propelled cleaner 1 travels, preferably at a number of positions that can be said to be substantially arbitrary positions. It is a process to register. To put it simply, it can be said that the self-propelled cleaner remembers the scene at each position in the room.
It is assumed that the instruction to create the location information database is predetermined as a voice message “Remember the room features” by the user, for example. When the voice recognition unit 135 receives this voice and recognizes it as an instruction, in response to this, the device control unit 11 sets the self-propelled cleaner 1 in various places in the room as a travel control unit 11a according to a rule determined in advance from the home position. To run. Then, the image capturing unit 133 is caused to capture a scene at each position. Further, the device control unit 11 extracts feature points from the captured images at the respective positions as the feature point extraction unit 11c. The room coordinates indicating where in the room the image was taken are detected by the device control unit 11 as the position detection unit 11b using encoder signals from the left wheel encoder 71L and the right wheel encoder 71R. The position detection unit 11b may further increase the accuracy of the room coordinates using the gyro sensor 20.

The device control unit 11 has, for a plurality of feature points extracted from an image captured at each position, feature point information indicating the position in the image from which each feature point is extracted, and room coordinates indicating the indoor position. Are stored in the positional information database storage unit 61a. The feature point information includes feature point coordinates indicating the position of each feature point in the image, and information (distance information) indicating the arrangement relationship of each feature point in the image.
Information in which feature point information created based on an image taken at a certain position in the room is associated with room coordinates representing the position is referred to as position information of the position. The device control unit 11 stores the position information at each photographing position in the position information database storage unit 61a. The process of creating position information at a plurality of positions in the room and storing it in the position information database storage unit 61a is referred to as “creation of a position information database”.

The procedure for creating the location information database in this embodiment is as follows.
When receiving an instruction to create a position information database while docked with the charging stand 201, the device control unit 11 controls the travel control unit 11a to travel the self-propelled cleaner 1 to various places in the room. Even when cleaning is performed, the self-propelled cleaner 1 travels through the room and cleans it. Therefore, the same traveling pattern may be applied when the position information database is created and when the cleaning is performed. . At this time, the running surface in the room is divided into vertical and horizontal matrices of a predetermined size, a feature point information is created by taking a picture for each section and extracting feature points, and the feature point information and room coordinates Is stored in the position information database.

FIG. 7 is an explanatory diagram showing an example of a room layout and sections where the self-propelled cleaner 1 travels in this embodiment. In a room 300 illustrated in FIG. 7, furniture such as a table 301, a sofa 303, a television 305, a bookshelf 307, and an air purifier 309 are placed. These furnitures are obstacles for the self-propelled cleaner 1, and for example, a place where the TV 305 is placed cannot run. A charging stand 201 is placed at the corner of the room (upper right corner in FIG. 7). A rectangular area separated by a chain line running vertically and horizontally in the room 300 indicates a unit of partition. Furniture is substantial, but the sections are virtual.
In this embodiment, the rule that the self-propelled cleaner 1 travels in the room when creating the position information database is basically to perform a zigzag operation while shifting the round-trip path at a predetermined interval (pitch). And In other words, the self-propelled cleaner 1 goes straight until an obstacle is detected in front of the traveling path, and when a wall that is an obstacle is detected, the self-propelled cleaner 1 avoids a collision with the obstacle (in this embodiment, the side). Turn 90 ° to the right. And if it advances only 30 cm of the predetermined pitch, it will turn 90 degrees to the right side and it will return. Then run according to the rule that you go straight until you find an obstacle again.

The position detection unit 11b detects room coordinates of the traveled route. If the traveling control unit 11a continues the basic zigzag operation and determines that the self-propelled cleaner 1 has traveled in the room to the corner on the opposite side of the home position, the traveling control unit 11a avoids an obstacle during the zigzag operation from the home position. Trying to travel in areas that did not travel. This is because when a floor area that has been folded back to avoid an obstacle remains, the vehicle travels in that area.
Note that the zigzag operation folding pitch when creating the position information database, in other words, the section pitch may be the same as or different from the folding pitch for cleaning.

FIG. 8 is an explanatory diagram showing a state in which the self-propelled cleaner 1 in this embodiment starts photographing from the home position and sequentially performs photographing in each section when creating the position information database. In FIG. 8, an arrow starting from the charging stand 201 indicates the traveling path of the self-propelled cleaner 1. Circles lined up along the arrows indicate positions where the device control unit 11 causes the imaging unit 133 to capture images in each section. As shown in FIG. 8, the device control unit 11 causes the self-propelled cleaner 1 to travel in a zigzag manner and causes the imaging unit 133 to capture a scene corresponding to the section every time the section is advanced. And the positional information linked | related with the room coordinate is stored in the positional information database storage part 61a.
The vertical and horizontal matrix-shaped sections shown in FIG. 8 are assumed to have a size of 30 cm in both vertical and horizontal directions. In that case, every time the self-propelled cleaner 1 goes straight 30 cm, the device control unit 11 causes the imaging unit 133 to photograph a surrounding scene at that position. The fact that the vehicle has traveled 30 cm straight is determined by comparing the room coordinates of the current position detected by the position detection unit 11b with the room coordinates of the previously photographed position. Then, the feature point extraction unit 11c extracts feature points from the captured image. Then, feature point information representing each coordinate value (feature point coordinates) in the image of the plurality of extracted feature points and the arrangement relationship of these feature points is created. In this specification, capital letters X and Y are used to represent the feature point coordinates. In FIG. 6A and FIG. 6B, the lower case x and y are used for the room coordinates indicating the position of the self-propelled cleaner 1 in the room, so that the distinction from the lower case is made clear.

When creating the feature point information relating to a certain image, the device control unit 11 creates distance information representing the arrangement relationship of the feature points in the image. In this embodiment, the distance information is data obtained by calculating the distance between two feature points included in the same image for all feature points. The feature point information includes feature point coordinates indicating the position of each feature point. Furthermore, distance information representing the distance between the feature point and all the feature points in the image is included. The feature point information is information obtained by combining feature point coordinates and distance information for all feature points in the same image. The position information is information that associates the room coordinates (x and y coordinate values of the shooting position) detected by the position detection unit with the feature point information, and the position information database stores the position information at each position in the room. It is.
The equipment control unit 11 stores the position information and room coordinates of the sections that can be traveled while the self-propelled cleaner 1 is traveling in the position information database storage unit 61a. However, if the current position is already within the section stored in the position information database storage unit 61a, duplicate shooting and position information are not stored. In this way, when the position information is stored in the position information database storage unit 61a for each section of the travelable area in the room, the device control unit 11 returns the self-propelled cleaner 1 to the charging base 201 to return to the home. Return to position.
The above is the outline of creating the position information database.

≪Learning the target position≫
The position information database created as described above is used to determine a target position where the self-propelled cleaner 1 should travel.
In this embodiment, the self-propelled cleaner 1 learns a target position by a user operation. Once the target position is learned, the self-propelled cleaner 1 can travel to the learned position simply by giving an instruction thereafter. The target position to be learned is not limited to one. A plurality of positions can be learned. Since the position learned by the self-propelled cleaner 1 can be a target position thereafter, it is called a candidate position.

At the stage where the candidate position is learned, the user takes the self-propelled cleaner 1 to the candidate position. In a state where the self-propelled type is placed on the floor surface of the candidate position, the self-propelled cleaner 1 is instructed to learn the position. For example, the user gives an instruction by voice saying “I ’m in front of the sofa, remember this place”.
When the voice recognition unit 135 receives this voice and recognizes it as an instruction, the self-propelled cleaner 1 captures a surrounding scene by the imaging unit 133 at that position in response to the instruction. Further, a plurality of feature points are extracted from the captured image, feature point information relating to each extracted feature point is created, distance information is created using the feature point information, and stored in the query storage unit 61b. . Here, the feature point extraction method, the feature point information, and the distance information creation procedure are the same as those in the creation of the position information database.
As the target position determination unit 11d, the device control unit 11 searches the position information database for position information having distance information most similar to the distance information stored in the query storage unit 61b. In this specification, this process is referred to as “position estimation”. That is, the nearest neighbor search process is performed based on the distance information at the candidate position. Note that an approximate nearest neighbor search may be performed instead of a strict nearest neighbor search in order to reduce the processing time and memory required for the search.

Then, the room coordinates (x and y coordinates) stored in the position information found from the position information database by the search process are determined as the target position, and the room coordinates are stored in the storage unit 61 as the room coordinates of the candidate position. Store.
Further, the learned candidate position is associated with a keyword that identifies the candidate position from other candidate positions. That is, the learned position is stored in the storage unit 61 in association with the room coordinates of the candidate position from the portion “Before the sofa”.
Alternatively, when the user simply gives an instruction “remember this position”, in response to this, an image is taken to extract a feature point, and a position information database is searched to determine room coordinates. After that, the device control unit 11 causes the notification unit 55 to output a voice message asking “Please say the name of this position!”. In response, when the user says, for example, “in front of the television”, the voice recognition unit 135 recognizes the voice. The device control unit 11 stores the recognized voice in the storage unit 61 in association with the room coordinates of the candidate position.

Thereafter, when the user utters a voice “clean the front of the sofa”, the voice recognition unit 135 receives the user's voice and recognizes the instruction to move to “in front of the sofa” and “clean”. The traveling control unit 11a retrieves and acquires the room coordinates associated with “in front of the sofa” among the candidate positions stored in the storage unit 61, and moves the self-propelled cleaner 1 to the position indicated by the acquired room coordinates. After running, the surroundings are cleaned, and control is performed to return to the home position when cleaning is completed.
Similarly, when the user's voice “clean the front of the television” is recognized, the device control unit 11 causes the self-propelled cleaner 1 to travel to the position in front of the television that has already been learned, and cleans the surroundings. , Control to return to the home position after cleaning.

≪Location information database creation process≫
Further detailed processing for creating the location information database will be described.
The position information database includes feature point information obtained from an image of a surrounding scene captured by the image capturing unit 133 at each position in the room and a relative position from the room coordinate origin detected by the position detecting unit 11b at the position where the image was captured. It is a database that is registered with the location linked.
9 to 11 are flowcharts showing processing executed when the device control unit 11 creates a position information database in this embodiment. FIG. 9 shows the overall processing, FIG. 10 shows the creation of feature point information, and FIG. 11 shows the creation of distance information.

A processing procedure will be described with reference to the flowchart of FIG.
When receiving an instruction from the user, the device control unit 11 creates a position information database in response to the instruction.
First, the device control unit 11 initializes variables used for the following processing. For example, the coordinate value zero indicating the home position is set in the temporary variables x and y storing the coordinate values of the room coordinates. Moreover, the initial value zero is set to the variable R representing the traveling direction of the self-propelled cleaner 1.
Here, the variable R takes a value from 0 to 3, and each value is
R = 0: the traveling direction is the positive direction of the y-axis of the room coordinates (upward direction in FIG. 6A)
R = 1: The traveling direction is the plus direction of the x axis of the room coordinates (the right direction in FIG. 6A).
R = 2: The traveling direction is the negative direction of the y-axis of the room coordinates (downward direction in FIG. 6A)
R = 3: The traveling direction is the minus direction of the x axis of the room coordinates (the left direction in FIG. 6A).
The positive and negative directions of the x and y coordinates are expressed as follows. For example, when an obstacle is detected in front of the vehicle during travel, the device control unit 11 turns the self-propelled cleaner 1 90 degrees to the right in order to avoid the obstacle. At the same time, 1 is added to the direction variable R (however, it is set to zero when R = 3).
Further, the device control unit 11 starts traveling in the room after turning the self-propelled cleaner 180 degrees (step S11).

While traveling in the room, the device control unit 11 checks whether an obstacle is detected in front of the traveling surface (step S13). If there is an obstacle (Y in Step S13), the self-propelled cleaner 1 is controlled to turn left or right by 90 ° immediately before that to perform zigzag running (Step S15). If there is no obstacle (N in step S13), the process continues to proceed (step S17), and proceeds as it is until it reaches the imaging position of a new section. During traveling, the device control unit 11 calculates a traveling distance and a traveling direction based on encoder signals obtained from the left wheel encoder 71L and the right wheel encoder 71R as the position detection unit 11b. Then, the self-propelled cleaner 1 is currently calculated in which position in the room coordinates based on the home position. Details of the processing for calculating the room coordinates will be described later.
(Step S19). When the shooting position of a new section is reached, it is checked whether or not position information has already been registered in the position information database at that position (step S21). If already registered (Y in step S21), the routine proceeds to step S29.

When the position information at the position is not registered (N in Step S21), the device control unit 11 causes the image capturing unit 133 to capture a scene at the position. Further, the feature point extraction unit 11c performs a process of extracting feature points from the captured image and creating feature point information (step S23). Detailed processing for creating feature point information is shown in FIG. 10 and will be described in detail with reference to FIG. The feature point information is represented as an array fp [] of a plurality of feature points extracted from one image.
Furthermore, the device control unit 11 creates distance information based on the created feature point information (step S25). Detailed processing for creating the distance information is shown in FIG. 11 and will be described in detail in the description of FIG. It is represented as a two-dimensional array ds [] [] corresponding to the point.

Then, the device control unit 11 associates the room coordinates of the position detected as the position detection unit 11b with the feature point information and the distance information as position information, and stores the position information in the position information database storage unit 61a (Step S1). S27).
Then, it is determined whether or not the position information has been registered in the position information database for all the sections that can travel in the room (step S29).
If there are still sections to be traveled (N in Step S29), the routine returns to Step S13, travels while avoiding obstacles, and repeats registration of position information in each section.
On the other hand, if it is determined that all the sections have been registered (Y in step S29), the process ends.

≪Calculation of room coordinates≫
Details of the room coordinate calculation described in step S19 will be described.
The position detection unit 11b sequentially detects a relative position from the home position of the self-propelled cleaner 1.
It is assumed that the self-propelled cleaner has traveled a distance D between the last time the position of the self-propelled cleaner 1 was detected and the current position was detected. This distance D is calculated based on encoder signals from the left wheel encoder 71L and the right wheel encoder 71R. If the count numbers of both encoder signals are equal, it is determined that the vehicle is traveling straight. If the signs are different but the absolute values are equal, it is determined that the self-propelled cleaner is turning. Here, the code | symbol of an encoder signal shall represent the rotation direction of a driving wheel.

  Furthermore, if there is a difference between the two, it is determined that the self-propelled cleaner 1 has traveled while changing the travel direction. It is possible to detect the direction of travel and the distance traveled according to the difference between the two. When the direction of travel changes slowly, the left and right rotational speeds are controlled to go straight, so for the sake of simplicity, in this embodiment, the travel pattern is only straight and 90 ° turn. That is, here, it is assumed that the self-propelled cleaner 1 travels straight along the x-axis or y-axis and travels a distance D.

The position detector 11b refers to the value of the variable R indicating the traveling direction in the previous and current position detection. As assumed, since the self-propelled cleaner 1 goes straight along the x-axis or the y-axis, the value of the variable R is equal between the previous time and the current time. When the room coordinates when the previous position was detected are (x _i , y _i )
When R = 0, the current room coordinate (x _j , y _j ) is calculated as a position obtained by adding D to the y coordinate value with respect to the previous room coordinate value. That is,
x _j = x _i , y _j = y _i + D
And
When R = 1, the position is calculated by adding D to the x coordinate value with respect to the previous room coordinate value.
x _j = x _i + D, y _j = y _i
In the case of R = 2, the position is calculated by subtracting D from the y coordinate value with respect to the previous room coordinate value.
x _j = x _i , y _j = y _i −D
In the case of R = 3, the position is calculated by subtracting D from the x coordinate value with respect to the previous room coordinate value.
x _j = x _i −D, y _j = y _i
When the self-propelled cleaner 1 is only turned, the current room coordinate (x _j , y _j ) is equal to the previous room coordinate value, and the variable R changes.
As a unit of the coordinate value of the room coordinates, for example, the size of one section of the vertical and horizontal matrix shown in FIG. 7 may be set as one unit. When the above coordinate values (x _j , y _j ) are measured in units of centimeters and the size of one section is 30 centimeters both vertically and horizontally, the coordinate values of the room coordinates are x _j and y _j of 30 cm each. The divided value.

≪Details of feature point information creation process≫
Next, details of the feature point information creation process shown in step S23 of FIG. 9 will be described with reference to the flowchart of FIG. FIG. 10 is a flowchart showing details of the process shown in step S23 of FIG.
In FIG. 10, the device control unit 11 as the feature point extraction unit 11c causes the imaging unit 133 to capture a surrounding scene (step S31). Then, the captured image is binarized into white and black (step S33). For the binarization processing, a known algorithm such as a P-tile method or a mode method may be applied. Note that the width (number of pixels in the X direction) of the image obtained by the binarization process is represented by “WIDTH”, and the height (number of pixels in the Y direction) of the image is represented by “HEIGHT”.

  Subsequently, variables used for the subsequent processing are initialized (step S35). Specifically, X = 0 is set to a variable indicating the number of pixels in the X direction of the feature point in the photographed image, in other words, the variable X representing the X coordinate. Similarly, Y = 0 is set to a variable indicating the number of pixels in the Y direction of the feature point in the image, in other words, a variable Y representing each Y coordinate. Further, all elements of the feature point array fp [] that is an array variable representing the feature points in one image are cleared. Also, i = 0 is set to the variable i of the pointer indicating the feature point array fp []. In this embodiment, the feature point information is represented as a feature point array fp [].

Subsequently, the feature point extraction unit 11c determines whether the 3 × 3 pixel matrix configured by the range of the feature point coordinates (X, Y) − (X + 2, Y + 2) corresponds to a pattern as a predetermined corner. Check whether or not. That is, a 3 × 3 pixel matrix composed of a pixel of interest (X + 1, Y + 1) in the center of the pixel matrix and a reference area of 8 pixels surrounding the pixel is collated with a corner pattern (matching is performed) (matching process in step S37) . The corner pattern is shown in FIG. 13 as a specific example and will be described later, but is prepared in advance as a 3 × 3 pixel matrix.
When the pixel matrix including the pixel of interest matches the corner pattern (Y in step S37), the following processing is performed. Assuming that the target pixel is a feature point, the feature point coordinate value of the target pixel (X + 1, Y + 1) is stored in the feature point array fp [] and the value of the pointer i is incremented by 1 (step S39).
On the other hand, when the pixel matrix including the pixel of interest does not match the corner pattern (N in step S37), storage in the feature point array fp [] is not performed.

Subsequently, the feature point extraction unit 11c increases the value of the variable X representing the number of pixels in the X direction by 1 (step S41), shifts the target pixel to the right side, and then 3 × 3 including the target pixel and the reference area. It is checked whether or not the pixel matrix has reached the right edge of the image (step S43). That is, it is checked whether or not the value of the variable X has reached “WIDTH” −2. Here, “−2” is based on the condition that a 3 × 3 pixel matrix having a width of three pixels does not protrude from the image.
If the right end has not been reached (Y in step S43), the routine returns to step S37 to check whether or not the pixel of interest after the shift corresponds to a corner pattern.
On the other hand, when the pixel matrix after the shift protrudes from the right end of the image (N in Step S43), the feature point extraction unit 11c increases the variable Y by 1 and shifts the target pixel downward. At the same time, X = 0 is set in the variable X, and the pixel of interest is moved to the left end of the shifted line (step S45). In this way, preparation for scanning the next line is made, but before that, it is confirmed whether the pixel matrix shifted downward before the start of scanning reaches the lower end of the image (step S47). That is, it is checked whether or not the value of the variable Y has reached “HEIGHT” −2. “−2” is based on the condition that a 3 × 3 pixel matrix having a width of three pixels does not protrude from the image.

If the lower end has not been reached (Y in step S47), the routine returns to step S37 to check whether or not the pixel of interest after the shift corresponds to a corner pattern.
On the other hand, when the pixel matrix after the shift protrudes from the lower end of the image (N in step S47), it is determined that the operation is completed, and the value of the pointer variable i at that time is stored in the variable “FPMAX”. The FPMAX and the feature point array fp [] provide feature point information corresponding to the image.

≪Details of distance information creation process≫
Next, details of the distance information creation processing shown in step S25 of FIG. 9 will be described using the flowchart of FIG. FIG. 11 is a flowchart showing details of the process shown in step S25 of FIG.
In FIG. 11, the device control unit 11 calculates the distance between two feature points in the same image for each combination of feature points.
First, initialize variables. First, h1 = 0 is set to the index variable h1 for referring to the first feature point (step S51). Further, each element of the two-dimensional distance information array ds [] [] storing the distance between the two feature points is cleared to zero.
Next, h2 = 0 is set to the index variable h2 for referring to the second feature point (step S53).

Subsequently, it is checked whether or not the values of the variables h1 and h2 are equal (step S55). Since h1 = h2 = 0 after initialization, both are equal (Y in step S55). If both are equal, the same feature point is pointed out, so the distance d is set to 0 (step S57). On the other hand, when h1 and h2 are not the same, the distance between the feature points fp [h1] and fp [h2] indicated by h1 and h2 is calculated (step S59). The distance d between the two feature points is defined as (X0, Y0) and (X1, Y1) as the feature point coordinates of the two feature points, respectively.
d = √ {(X1-X0) ^ 2 + (Y0-Y1) ^ 2}
Is required. That is, it can be calculated as a Euclidean distance in a two-dimensional plane.

When the distance is obtained in the above step S57 or S59, the obtained distance d is stored in the distance information array ds [h1] [h2] (step S61).
For example, one case eye elements feature points and the feature point distance is fourth element is of the _{d 14, h1 = 0, h2} = 3 , and therefore the distance information sequence ds [0] to [3] d ₁₄ is stored. That is,
ds [0] [3] = d ₁₄
The distance _{d 41} of the fourth characteristic points are the elements and the feature point is a first element, since the h1 = 3, h2 = 0, the _{ds [3] [0] =} d 41 . Here, since ds [0] [3] and ds [3] [0] are the same information, either one may be omitted. Here, in order to simplify the description, it is assumed that distances are calculated for all combinations of two feature points and stored in the distance information array ds [] [].

Next, the device control unit 11 increases the index h2 referring to the second feature point by one (step S63), and checks whether h2 has referred to all the feature points (step S65). . That is, it is checked whether or not the value of h2 has reached FPMAX. Here, the variable FPMAX stores the number of feature points extracted from the image when the feature point information is created in FIG.
If h2 does not reach FPMAX, that is, if h2 <FPMAX (Y in step S65), the routine returns to step S55 to calculate the distance between two new feature points, and stores it in the distance information array ds [] []. Store.
On the other hand, if h2 reaches FPMAX (N in step S65), the routine proceeds to step S67 and increments the index variable h1 that refers to the first feature point by one (step S67). It is checked whether or not all feature points have been referred to (step S69).

If h1 has not reached FPMAX, that is, if h1 <FPMAX (Y in step S69), the routine returns to step S53. Then, h2 is initialized to zero, and the distance from all the feature points in the image for the new first feature point is calculated and stored in the distance information array ds [] []. That is, for each new h1, the distances of combinations of h2 from 0 to FPMAX-1 are calculated and stored in the distance information array ds [] [].
On the other hand, if h1 reaches FPMAX in step S69 (N in step S69), the process ends.

≪Location information database data structure≫
FIG. 12 is an explanatory diagram showing the data structure of the location information database created in this embodiment.
As shown in FIG. 12, the position information database includes position information in each room coordinate. In the example of FIG. 12, position information at n different room coordinate positions is stored in the position information database. Each piece of position information includes room coordinate values x and y representing the position and a feature point array (that is, feature point information) relating to an image photographed at the position.
The feature point information is an array of feature points extracted from one image. In the example of FIG. 12, m feature points are extracted from one image.
One element corresponding to each feature point in the feature point array includes feature point coordinates (X, Y) indicating the position of the feature point in the image, and the feature point as the first feature point in the image. It consists of distance information ds [] [] indicating the distance to all feature points. The distance information related to the h-th feature point among the m feature points is represented by ds [h] [k]. However, k takes m values from 0 to (m−1). Therefore, ds [h] [k] is composed of elements in which h is fixed and k is m.

≪Example of feature point pattern≫
FIG. 13 is an explanatory diagram showing pixel patterns detected as feature points in this embodiment. In FIG. 13, each pixel pattern represents a 3 × 3 pixel matrix, and each pixel in the pixel matrix represents binarized white or black.
In this embodiment, the 3 × 3 pixel matrix area of the binarized image is matched with a predetermined corner pixel pattern. FIG. 13 shows pixel patterns of predetermined corners by type.
In FIG. 13, the eight pixel patterns shown in FIG. The eight pixel patterns shown in (b) indicate detection patterns for obtuse corners. Further, the eight pixel patterns in (c) show detection patterns for acute corners. When the pixel pattern matches any of these pixel patterns, the center pixel of the 3 × 3 pixel matrix is set as the feature point coordinates.

On the other hand, the eight pixel patterns shown in (d) are edge patterns. Since it is not a corner, it is not extracted as a feature point.
However, this is only an example, and an aspect in which edges as well as corners are extracted as feature points is also conceivable. Further, the 3 × 3 pixel matrix is only an example, and a mode in which a pixel matrix of another size is applied for matching is also conceivable.
Note that SIFT, SURF, ORB, AKAZE, and the like are known as methods for extracting feature points from an image, and these known algorithms may be applied.

  FIG. 14 is an explanatory diagram visually showing a position where a feature point is extracted from an image obtained by photographing a table as an example of an object in this embodiment. As shown in FIG. 14, after binarizing the table image, the pixel of interest in the image is shifted vertically and horizontally to find a point where the 3 × 3 pixel matrix matches the corner detection pattern of FIG. Go. For example, the lower end of the left leg on the front side of the table matches the detection pattern at the right-angled corner, and thus is extracted as a feature point. In addition, the lower end of the right leg on the front side of the table matches the pattern of the obtuse corner and is extracted as a feature point. On the other hand, an edge pattern exists at the front center of the table, but it is not a feature point because it is not a corner. In this way, a plurality of feature points that match the corner detection pattern are extracted from the table image.

≪Position estimation process≫
Next, details of position estimation in this embodiment will be described.
It should be noted that, since the data search in the “position estimation” process takes a calculation time, the calculation process of the feature points may be performed while the self-propelled cleaner 1 is docked with the charging stand 201.
In addition, although the imaging unit 133 performs shooting at a position to be learned in advance, the position information database creation process may be performed while cleaning. Then, data creation for the “position estimation” process may be performed after the creation of the position information database and the completion of the database.

FIG. 15 is a flowchart showing the overall process of position estimation performed by the device control unit 11 in this embodiment.
For example, in response to a user instruction requesting position estimation such as “Let's remember this place, in front of the sofa”, the device control unit 11 executes the following process as the target position determination unit 11d.
First, the image capturing unit 133 is caused to capture a surrounding scene at that position. Feature point information is created from the captured image (step S101). The feature point information creation process is the same as the process described with reference to FIG. 10 and thus will not be described in detail. Create feature point information. Then, the contents of the feature point array fp [] and the value of the number of feature points are stored in the query storage unit 61b.

Subsequently, the target position determination unit 11d creates distance information (step S103). Although detailed description is omitted because it is the same as the processing of FIG. 11, the distance between the feature points is calculated for all combinations of the two feature points based on the created feature point array fp [], and the distance information array Store in ds [] []. Then, the contents of the distance information array ds [] [] are stored in the query storage unit 61b.
Furthermore, the target position determination unit 11d initializes variables used in the following processing (step S105). First, j = 0 (zero) is set in an index variable j for referring to the position information database so that the first position information among a plurality of position information stored in the position information database is indicated. Also, the temporary variable smax that stores the maximum score is cleared by setting smax = 0 (zero). Then, k = −1 is set to the index variable k for referring to the maximum score. Further, the comparison feature point array cfp [], the comparison feature point variable CFPMAX, and the comparison distance information array cds [] [] are cleared.

Next, the target position determination unit 11d uses the feature point information and the distance information captured at the candidate positions and captured at the steps S101 and S103 to sequentially match all the position information stored in the position information database. Execute the process.
First, it is checked whether or not the index j referring to the position information database is less than a value indicating the number of position information stored in the position information database, RMAX (step S107). Since the initial value of the variable j is zero, the determination in step S107 in that case is Yes, and the routine proceeds to step S109. On the other hand, when the determination is No, that is, when the matching is completed for all the position information, the process is terminated.

If the determination in step S107 is Y, the target position determination unit 11d extracts one piece of position information from the position information database using the index variable j (step S109). That is, the position information is to be compared with information related to the target position stored in the query storage unit 61b. Then, the feature point array related to the extracted position information is copied to the comparison feature point array cfp [], and the number of feature points is copied to CFPMAX. Further, the distance information related to the position information is copied to the comparison distance information array cds [] [].
Note that the position information to be compared is sequentially changed by the loop process from step S117 to step S107, and loops until all position information in the position information database is compared.

Subsequently, the target position determination unit 11d calculates a score s indicating the degree of matching using the feature point arrays fp [], cfp [], the distance information arrays ds [] [], and cds [] [] (step S111). fp [] is a feature point array of the query storage unit 61b, and cfp [] is a feature point array in which a comparison target is stored. ds [] [] is a distance information array of the query storage unit 61b, and cds [] [] is a distance information array in which a comparison target is stored. Details of the procedure for calculating the score s will be described with reference to FIG.
The target position determination unit 11d checks whether or not the calculated score s is greater than smax (step S113). When the score s is larger than smax (Y in step S113), the score s is set in the variable smax. Also, the index variable k for referring to the maximum score is updated with j (step S115). That is, the highest score so far is stored in the variable smax, and an index referring to the position information having the score is stored in the variable k.
Then, after incrementing the index variable j by one (step S117), the routine returns to step S107.

As described above, all the pieces of position information stored in the position information database are matched, the position information index that gives the highest score is stored in the variable k, and the score is stored in smax. finish.
When the processing is completed, the variable k indicates position information having a feature most similar to the feature at the target position stored in the query storage unit 61b. If similar position information is not found, the variable k remains the initial value k = -1. When the variable k is not −1, the position information referred to by the variable k, that is, the room coordinates (x, y) of the position information corresponding to the kth record of the position information database is the estimated target position.

≪Matching process≫
Details of the matching process executed in step S111 in FIG. 15 will be described.
FIG. 16 is a flowchart showing the procedure of matching processing in this embodiment.
In FIG. 16, first, the target position determination unit 11d sets a variable s for storing a score to s = 0 and clears it (step S121).
In calculating the score, a reference feature point (hereinafter simply referred to as a reference point) for evaluating the positional relationship of the feature point coordinates from the feature points related to the feature point array fp [] stored in the query storage unit 61b. Processing to determine is performed (step S123). That is, when the matching between the feature point array fp [] and the feature point array cfp [] storing the position information to be compared is examined, a feature point (reference point) to be used as a matching reference is determined.

  In this embodiment, the matching is determined based on whether there is at least one triangle in the feature point array cfp [] that is congruent with the three points in the feature point array fp []. Six points constituting a congruent triangle pair (three points paired with three points) are called reference points. Two points out of both three points constituting the reference point are selected, and a score representing the degree of matching is calculated using the two points (matching reference point) and the feature point indicated by the other index variable q1. The details will be described with reference to FIG. 17. In step S123, the combination of the three points in the feature point array fp [] stored in the query storage unit 61b and the feature of the location information to be compared in the location information database. A combination of three points in the point array cfp [] is sequentially compared. Then, one of the three points corresponding to the arrangement relationship (three points with congruent triangles having three points as vertices) is found and set as the three reference points. Of the three points found, index variables indicating two points on the fp [] side (matching reference point) are p1 and p2, and index variables indicating two points on the cfp [] side (matching reference point) are p3 and p4. Set to. fp [p1] corresponds to cfp [p3], and fp [p2] corresponds to cfp [p4].

When two matching reference points in the feature point array fp [] are determined, the target position determination unit 11d then clears the index variable q1 used for referring to the feature point array fp [] by setting q1 = 0. (Step S125). Subsequently, the target position determination unit 11d checks whether or not the value of the index variable q1 is less than the number of feature points FPMAX constituting the feature point array fp [] (step S127). That is, it is checked whether or not the feature point coordinates have been checked for all feature points stored in the query storage unit 61b.
If the variable q1 is less than FPMAX (Y in step S127), a process for searching for a corresponding point (step S129) is performed. The details of the corresponding point search processing will be described with reference to FIG. 19, and it is checked whether or not there is a corresponding point related to the corresponding triangle in the feature point array cfp [] to be compared. As a result, if there is a corresponding point (Y in step S131), the score s is increased by 1 (step S133). If there is no corresponding point (N in step S131), the score s is left as it is. As will be described with reference to FIG. 19, when three points having the same distance relationship are not found in step S129, the value of the variable m1 is set to m1 = −1 and the process returns.

Thereafter, the target position determination unit 11d increases the index variable q1 by 1 (step S135). The routine then returns to step S127 and loops.
When the corresponding points are searched for all the feature points constituting the feature point array fp [] as described above (N in step S127), the process is terminated. At the time when the processing is completed, the score s is congruent among a plurality of sets of triangles having three vertices of the other feature points in the same image as the two matching reference points determined in step S123. This represents how many triangles (three corresponding points) exist. As the score s increases, the feature point arrangement relationship between the feature point array fp [] and the comparison target feature point array cfp [] is more similar.

≪Determination of reference point≫
FIG. 17 is a flowchart showing details of the reference point determination process shown in step S123 in the matching process shown in FIG.
In FIG. 17, the target position determination unit 11d clears each of the variables m1, m2, and m3 used for processing by setting zero (step S141). m1, m2, and m3 are all index variables used for referring to the feature point array fp [], and change in different values while taking different values so that the feature points to be referenced do not overlap in the following processing.
Next, the target position determination unit 11d sets m2 = m1 + 1 to the index variable m2 (step S143). Step S143 is in the loop and is repeatedly processed. In the first time, m1 is an initial value (zero), and m2 is one larger value, m2 = 0 + 1 = 1.

Subsequently, the target position determination unit 11d sets m3 = m2 + 1 to the index variable m3 (step S145). Step S145 is in the loop and is repeated, but for the first time, m2 is 1 and m3 is one larger value, m3 = 1 + 1 = 2.
Subsequently, the target position determination unit 11d executes a reference point search process (step S147). That is, the feature point arrays fp [m1], fp [m2], and fp [m3 that are referred to using the current m1, m2, and m3 among the feature point arrays fp [] stored in the query storage unit 61b. ] Is obtained. If these are assumed to be one of the three reference points, a search is made as to whether or not a feature point (the other of the reference points) having the same distance relationship exists in the position information to be compared in the position information database. Details of the process of searching for the corresponding three points will be described with reference to FIG.

As a result of the search, if there is a reference point, that is, a pair of three points having the same distance relationship (Y in step S149), the process ends. If not (N in step S149), the following process is looped. To do. As will be described with reference to FIG. 18, if no reference point is found in step S149, the value of variable p1 is set to p1 = −1 and the process returns. On the other hand, when a corresponding reference point is found, index variables that refer to 2 points out of both 3 points forming a pair and 4 points as a whole pair are set to p1 to p4.
First, the target position determination unit 11d increases m3 having the largest value among the three index variables m1, m2, and m3 by one (step S151), and checks whether the number of feature points is less than FPMAX (step S151). S153). That is, it is checked whether the index points within the feature point array.
If within the feature point arrangement range (Y in step S153), the routine returns to step S147. Then, the feature point referred to using the updated m3, m1, and m2 is assumed to be one of the reference points, and a search is performed as to whether or not the feature point having the same distance relationship exists in each location information in the location information database. To do.

On the other hand, if the variable m3 is not within the range of the feature point array (N in step S153), the variable m2 is increased by 1 this time, assuming that the range that m3 can take is covered (step S155). Then, it is checked whether m2 is less than (FPMAX-1) (step S157). That is, it is checked whether the value of m2 is within a range that can be taken as an index. In order not to overlap with m3, the maximum value that m2 takes is one less than FPMAX, which is the maximum value of m3.
If m2 is within the range of possible values (Y in step S157), the routine returns to step S145, and sets the value of m3 for the updated m2. When the feature point referred to using m1, m2, and m3 is assumed to be one of the reference points, a search is performed as to whether or not a feature point having the same distance relationship exists in each location information in the location information database.

  On the other hand, if it is not within the range of values that the variable m2 can take (N in step S157), then the variable m1 is incremented by 1 (step S159), assuming that the range that m2 can take is covered. Then, it is checked whether m1 is less than (FPMAX-2) (step S161). That is, it is checked whether the value of m1 is within a range that can be taken as an index. Note that the maximum value taken by m1 is two less than FPMAX, which is the maximum value of m3, so as not to overlap with m2 and m3.

If m1 is within the range of possible values (Y in step S161), the routine returns to step S143, and sets the values of m2 and m3 for the updated m1. When the feature point referred to using m1, m2, and m3 is assumed to be one of the reference points, a search is performed as to whether or not there is a feature point having the same distance relationship in the position information database.
On the other hand, if it is not within the range of values that the variable m1 can take (N in step S161), it is determined that the range that the m1 can take is covered. Then, in order to indicate that there are no corresponding three points even when combining the feature points of the feature point array fp [] in all possible ranges of m1, m2 and m3, m1 = -1 is set to the variable m1. (Step S163), the reference point extraction process ends.
As described above, in the process of FIG. 17, three points having the same distance relationship as the three points in the feature point array fp [] stored in the query storage unit 61b are compared with the position information cfp [] (step of FIG. 15). If the variable m1 is not found in (obtained in S109), the variable m1 takes the value of m1 = -1. When the corresponding reference point is found, the value related to the index of the feature point coordinates indicating the reference point of the feature point array fp [], cfp [] is temporarily one of the variables m1, m2, m3, n1, n2, and n3. Stored in Then, as will be described in detail later with reference to FIG. 18, those values finally enter variables p1, p2, p3 and p4.

≪Validity of checking correspondence with distance relation of 3 points≫
Here, the correspondence is checked using the distance relationship between the three points in consideration of the fact that the feature points stored in the two images to be compared are rotated or moved horizontally and vertically. .
That is, when the self-propelled cleaner 1 travels on the floor and changes its position, the image captured by the imaging unit 133 changes corresponding to the change of position. In this embodiment, when the imaging unit 133 captures a scene of the upper part and its surroundings, feature points extracted from an object on a ceiling surface parallel to the floor surface or an object near the ceiling surface are translated. And undergo only changes in rotation.
If the change of the image accompanying the change of the position of the self-propelled cleaner 1 is regarded as a geometric transformation, the ceiling surface can be considered as a geometric transformation only for rotation and translation, that is, Euclidean transformation. The distance between the two points does not change even if the image is subjected to Euclidean transformation. That is, the distance between the two points is known as an invariant (a quantity that does not change even when subjected to a geometric return) for the translational (translational) and rotational geometric transformations.

For images of objects located away from the ceiling surface, in addition to rotation and translation as the self-propelled cleaner 1 moves, reduction, enlargement, shear distortion (distortion in which a rectangle becomes a parallelogram), Projection distortion (distortion in which a rectangle becomes trapezoid) is affected. However, if the section is fine, the positions of the target section and the nearest section are close, the difference between the images taken at those positions is small, and enlargement / reduction, shearing distortion, or projection distortion can be ignored. However, the influence of rotation due to the difference in the direction of the self-propelled cleaner 1 cannot be ignored even at the same position.
As described above, a corresponding image may be searched in consideration of Euclidean transformation with rotation and translation. Therefore, a search using the distance relationship between feature points, which is an invariant for Euclidean transformation, is effective.

≪Search for corresponding reference point-When determining reference point≫
FIG. 18 is a flowchart showing the reference point search process shown in step S147 of FIG. 17, that is, the process of searching for the corresponding three points.
In FIG. 18, the target position determination unit 11d clears index variables n1, n2, and n3 that refer to the comparison feature point array (step S171). That is, n1 = 0, n2 = 0, and n3 = 0 are set in the variables n1, n2, and n3, respectively. Also, the four index variables p1, p2, p3, and p4 that store two points as reference points among the three points that make a pair are cleared.

Next, the target position determination unit 11d sets n2 = n1 + 1 to the index variable n2 (step S173). Step S173 is in the loop and is repeatedly processed, but n1 is the initial value (zero) at the first time, and n2 is one larger value, n2 = 0 + 1 = 1.
Further, the target position determination unit 11d sets n3 = n2 + 1 to the index variable n3 (step S175). Step S175 is in the loop and is repeatedly processed. In the first time, n2 is 1, n3 is one larger value, and n3 = 1 + 1 = 2.

Subsequently, the target position determination unit 11d includes the three feature points of the feature point arrays fp [m1], fp [m2], and fp [m3], the comparison feature point arrays fp [n1], fp [n2], and It is examined whether or not the three feature points of fp [n3] have the same distance relationship. Specifically, it is examined whether or not each of the three feature points satisfies any of the following distance condition 1, distance condition 2, or distance condition 3.
Here, the distance information array based on the feature point array fp [] is ds [] [], the comparison distance information array based on the comparison feature point array cfp [] is cds [] [], and the distances are compared. Let DLIMIT be the permissible error and denote the absolute value of X by ABS (X).
When the allowable error DLIMIT is described, the position where the position information is registered and the candidate position are not necessarily the same position. Therefore, it is necessary to allow a certain amount of error when collating captured images. However, since this allowable error affects the recognition accuracy, it should be set with sufficient consideration.

The distance condition 1 is
ABS (ds [m1] [m2] −cds [n1] [n2]) <DLIMIT and ABS (ds [m2] [m3] −cds [n2] [n3]) <DLIMIT and ABS (ds [m3] [M1] -cds [n3] [n1]) <DLIMIT
It is. In other words, the distance relationship between the three points, in other words, the difference in the lengths of the sides of the triangle with the three points as vertices are all within a tolerance, and the triangles are substantially congruent. .

The distance condition 2 is
ABS (ds [m1] [m2] -cds [n2] [n3]) <DLIMIT and ABS (ds [m2] [m3] -cds [n3] [n1]) <DLIMIT and ABS (ds [m3] [M1] -cds [n1] [n2]) <DLIMIT
It is. This indicates that although the correspondence between the three points is different from the “distance condition 1”, the distance relations are all within the allowable error range.

The distance condition 3 is
ABS (ds [m1] [m2] -cds [n3] [n1]) <DLIMIT and ABS (ds [m2] [m3] -cds [n1] [n2]) <DLIMIT and ABS (ds [m3] [M1] -cds [n2] [n3]) <DLIMIT
It is.
This indicates that although the correspondence between the three points is different from “Distance Condition 1” and “Distance Condition 2”, the distance relations are all within the allowable error range.

The target position determination unit 11d first checks whether the distance condition 1 is satisfied (step S177). If the distance condition 1 is satisfied (Y in step S177), p3 = n1 is set in the index variable p3, and similarly p4 = n2 is set in p4 (step S201). p3 and p4 are index variables for referring to the comparison feature point array cfp [].
Further, the target position determination unit 11d sets p1 = m1 to the index variable p1 for referring to the feature point array fp [] stored in the query storage unit 61b, and similarly sets p2 = m2 to p2 ( Step S207). Then, the process ends.

On the other hand, if the distance condition 1 is not satisfied (N in step S177), the routine proceeds to step S179 and checks whether the distance condition 2 is satisfied. If the distance condition 2 is satisfied (Y in step S179), p3 = n2 is set to the index variable p3 for referring to the comparison feature point array cfp [], and similarly, p4 = n3 is set to p4. (Step S203). Thereafter, the routine proceeds to step S207 described above, sets p1 = m1 to the index variable p1, and similarly sets p2 = m2 to p2, and ends the process.
If the distance condition 2 is not satisfied (N in step S179), the routine proceeds to step S181 and checks whether the distance condition 3 is satisfied. If the distance condition 3 is satisfied (Y in step S181), p3 = n3 is set to the index variable p3 for referring to the comparison feature point array cfp [], and similarly p4 = n1 is set to p4. (Step S205). Thereafter, the routine proceeds to step S207 described above, sets p1 = m1 to the index variable p1, and similarly sets p2 = m2 to p2, and ends the process.

As described above, when a pair of reference points having the same distance relationship is found, the feature point array fp [] stored in the query storage unit 61b corresponds to the comparison feature point array cfp [] in the position information database. Set the index variable to refer to the point and finish the process.
On the other hand, when none of the distance conditions 1 to 3 is satisfied (N in Step S181), the target position determination unit 11d performs the following process. First, among the three index variables n1, n2 and n3, n3 having the largest value is incremented by one (step S183). Then, it is checked whether or not the number of feature points is less than CFPMAX (step S185). That is, it is checked whether the index points within the range of the comparison feature point array.
If it is within the range of the comparison feature point array (Y in step S185), the routine returns to step S177. When the comparison feature point referred to using the updated n3, n1, and n2 is assumed to be one of the reference point pairs, the other reference point pair that satisfies any of the distance conditions 1 to 3 is queried. A search is made as to whether or not the feature points are stored in the storage unit 61b.

On the other hand, if the variable n3 is not within the range of the comparison feature point array (N in step S185), the variable n2 is increased by 1 (step S187), assuming that the range that n3 can take is covered. Then, it is checked whether n2 is less than (CFPMAX-1) (step S189). That is, it is checked whether the value of n2 is within a range that can be taken as an index. In order not to overlap with n3, the maximum value taken by n2 is one less than CFPMAX which is the maximum value of n3.
If n2 is within the range of possible values (Y in step S189), the routine returns to step S175, and sets the value of n3 for the updated n2. When the feature point referred to using n1, n2, and n3 is assumed to be one of the reference point pairs, the other reference point pair that satisfies any one of the distance conditions 1 to 3 is stored in the query storage unit 61b. It is searched whether it exists in the arranged feature point array.

On the other hand, if it is not within the range of values that the variable n2 can take (N in step S189), the variable n1 is increased by 1 (step S191), assuming that the range that n2 can take is covered. Then, it is checked whether n1 is less than (FPMAX-2) (step S193). That is, it is checked whether the value of n1 is within a range that can be taken as an index. It should be noted that the maximum value taken by n1 is two less than FPMAX which is the maximum value of n3 so as not to overlap with n2 and n3.
If n1 is within the range of possible values (Y in step S193), the routine returns to step S173 and sets the values of n2 and n3 for the updated n1. When the feature point referred to using n1, n2, and n3 is assumed to be one of the reference point pairs, the other reference point pair that satisfies any one of the distance conditions 1 to 3 is stored in the query storage unit 61b. It is searched whether it exists in the arranged feature point array.

On the other hand, if it is not within the range of values that the variable n1 can take (N in step S193), it is determined that the range that n1 can take is covered.
The target position determination unit 11d indicates that there are no corresponding three points even when combining the feature points of the feature point array fp [] in all the ranges that n1, n2, and n3 can take. -1 is set (step S195), and the reference point pair extraction process is terminated.
As described above, when the reference point pair is not found in the process of FIG. 18, the variable p1 takes the value of p1 = -1. On the other hand, when a reference point pair is found, index variables that refer to the corresponding points are set to p1 to p4. Among the reference points, the feature point of fp [p1] corresponds to the feature point of cfp [p3], and the feature point of fp [p2] corresponds to the feature point of cfp [p4].

≪Corresponding point search process≫
Next, details of the corresponding point search processing shown in step S129 of FIG. 16 will be described.
FIG. 19 is a flowchart showing the corresponding point search processing shown in FIG. As shown in FIG. 19, the target position determination unit 11d first sets q2 = 0 to the index variable q2 for referring to the comparison feature point array cfp [] and clears it (step S211).
Then, it consists of three feature points consisting of the feature point related to the feature point array fp [q1] and any other two points, and a feature point related to the comparison feature point array cfp [q2] and any other two points. It is examined whether the three feature points meet any of the following distance conditions A, B, and C (step S213).

The distance condition A is
ABS (ds [p1] [p2] -cds [p3] [p4]) <DLIMIT and ABS (ds [p2] [q1] -cds [p4] [q2]) <DLIMIT and ABS (ds [q1] [P1] -cds [q2] [p3]) <DLIMIT
It is. This indicates that the lengths of the sides of the triangle with the three vertices are equal within a range that is less than the allowable error, and the triangles are substantially congruent.

The distance condition B is
ABS (ds [p1] [p2] -cds [p4] [q2]) <DLIMIT, and ABS (ds [p2] [q1] -cds [q2] [p3]) <DLIMIT, and ABS (ds [q1] [P1] -cds [p3] [p4]) <DLIMIT
It is. This indicates that the three-point correspondence is different from the “distance condition A”, but the triangles are substantially congruent.

The distance condition C is
ABS (ds [p1] [p2] -cds [q2] [p3]) <DLIMIT and ABS (ds [p2] [q1] -cds [p3] [p4]) <DLIMIT and ABS (ds [q1] [P1] -cds [p4] [q2]) <DLIMIT
It is. This indicates that the triangles are substantially congruent although the correspondence between the three points is different from “distance condition A” and “distance condition B”.
That is, the three feature points fp [p1], fp [p2], fp [q1] stored in the query storage unit 61b and the three feature points cfp [p3], cfp [related to certain position information in the position information database. It is determined whether or not p4] and cfp [q2] have the same distance relationship. Here, fp [p1], fp [p2], cfp [p3], and cfp [p4] are matching reference points determined in the process of step S123.

If any one of the distance conditions A, B, and C is satisfied (Y in step S213), the process ends.
On the other hand, when none of the distance conditions A, B, and C are satisfied (N in Step S213), the target position determination unit 11d first increases the index variable q2 by 1 (Step S215), and q2 is less than the feature point number CFPMAX. It is checked whether or not (step S217). This is a determination relating to loop processing for checking all combinations of coordinates of the feature point array cfp [].
If q2 is less than CFPMAX, the routine returns to step S213 and checks whether the updated index variable q2 meets any of the distance conditions A, B, and C.

On the other hand, when q2 reaches CFPMAX, all combinations of coordinates in the feature point array cfp [] are checked, but it is determined that none of the distance conditions A, B, and C is met. In order to express it, q2 = −1 is set to the index variable q2 (step S219), and the process is terminated.
The above is the corresponding point search processing.

≪Visual description of feature point matching≫
The matching process of the feature points of the feature point array fp [] and the comparison feature point array cfp [] shown in FIGS. 16 to 19 will be described visually.
FIG. 20A and FIG. 20B are explanatory diagrams visually showing a determination process of a reference point used for feature point matching in this embodiment.
As illustrated in FIG. 20A, the target position determination unit 11d first selects three feature points from the feature point array fp [] and the comparison feature point array cfp [] when determining the reference point. There are a plurality of combinations for selecting three feature points from fp [], and there are also a plurality of combinations for selecting three feature points from cfp [].
The target position determination unit 11d calculates the distance between the two feature points out of the three points, that is, the lengths of the three sides of the triangle having the three points as vertices. Then, a triangle having the same length on three sides is searched from fp [] and cfp []. In the example shown in FIG. 20A, fp [] consists of 15 feature points, and cfp [] consists of 12 feature points. Three feature points each indicated by an asterisk are congruent triangles, and a circle indicates another feature point.

When a congruent triangle is found, two of the three vertices (feature points) are selected as matching reference points, and the distance relationship between the two matching reference points and the other feature points is examined. Then, a search is made from fp [] and cfp [] for two matching reference points and another one feature point having the same distance relationship, that is, a congruent triangle (reference point pair). The number of congruent triangles found is calculated as a score.
In FIG. 20B, if three reference points are excluded from the 15 feature points of fp [], the number of feature points is 12. There are twelve triangles formed by these 12 matching reference points. On the other hand, if three reference points are excluded from the 12 feature points of cfp [], the number of feature points is 9. There are nine triangles formed by these nine matching reference points. Therefore, the number of congruent triangle combinations among the 12 triangles related to fp [] and the 9 triangles related to cfp [] exists, and the number is set as the matching score s. In the case of FIG. 20B, since there are nine OK feature points, the score s is s = 9/12.
In addition, a certain amount of error is allowed when comparing the lengths of the sides of the triangle. In the description of the flowchart, the value of this allowable error is DLMIT. Since the size of DLIMIT affects recognition accuracy, it should be appropriately determined by the designer according to camera characteristics and shooting intervals.
As described above, since the distance between the two points is invariant to the Euclidean transformation with rotation and translation, even if the direction of the self-propelled cleaner 1 is different between fp [] and cfp [], it is highly accurate. Matching can be judged.

  FIG. 21 is an explanatory diagram visually showing an image search based on feature point matching in this embodiment. If the position (viewpoint) is different, the arrangement relationship of the feature points changes even if the same object is viewed. The position of the viewpoint (photographing position) can be determined based on the matching relationship between the image to be matched and the arrangement relationship of the feature points of the image at each position, and the image having the most similar arrangement relationship to the image to be matched. Specifically, for each feature point, the distance from two feature points (matching reference points) that serve as a matching reference is calculated. If the relationship between the matching target and the distance is close, OK is determined, and if not, NG is determined. As a score, the number of feature points of OK may be used as a score.

  FIG. 22 is an explanatory diagram visually showing the concept of the score to be calculated in this embodiment. In FIG. 22, the feature point array fp [] is created from the image of the place to be learned as the target position. Images at each position in the position information database are stored in the comparison feature point array cfp [] when matching is performed. When three points corresponding to the distance relationship are found in both, the score s is calculated. The denominator of the score s is obtained by subtracting the number 3 of reference points indicated by asterisks in FIG. 22 from the number of feature points in the feature point array fp []. The branch of the score s is the number of overlapping feature points among the other feature points indicated by ◯ when both reference points (indicated by asterisks) are overlapped.

≪Example of matching room images≫
FIG. 23 is an explanatory diagram showing the arrangement and shooting positions of the shot rooms in this embodiment. The self-propelled cleaner stores the position information obtained by photographing the surrounding scene at each position of the vertical and horizontal matrix sections in the position information database storage unit. (5) A description will be given with images taken at five positions.
In FIG. 23, illumination A, illumination B, and an air conditioner whose frames are indicated by broken lines are objects arranged on or near the ceiling. Other objects are placed on the floor, but they are all quite high.

In this embodiment, the imaging unit 133 includes a fish-eye lens that can shoot the upper and surrounding scenes at once, and shoots upward.
FIG. 24A is an explanatory diagram illustrating an example of an image photographed at the candidate position in the room illustrated in FIG. 23 in this embodiment. It is an image of the position which wants self-propelled cleaner 1 to learn as a target position.
FIG. 24B is an explanatory diagram illustrating an example of an image captured at the position (1) illustrated in FIG.
24A and 24B show feature points extracted from both images. As for the feature points, the reference point is indicated by an asterisk and the other points are indicated by ◯. Further, feature points having the same distance relationship are indicated by solid lines, and feature points having the distance relationship not indicated are indicated by chain lines. The score relating to the matching between the two is s = 4/8.

FIG. 25A is an explanatory diagram showing an image taken at the same position as FIG. 24A.
FIG. 25B is an explanatory diagram illustrating an example of an image captured at the position (2) illustrated in FIG.
The notation method of the feature points is the same as in FIGS. 24A and 24B. The matching score is s = 6/8.
FIG. 26A is an explanatory diagram showing an image taken at the same position as FIG. 24A.
FIG. 26B is an explanatory diagram illustrating an example of an image captured at the position (3) illustrated in FIG.
The notation method of the feature points is the same as in FIGS. 24A and 24B. The matching score is s = 5/8.
FIG. 27A is an explanatory diagram showing an image taken at the same position as FIG. 24A.
FIG. 27B is an explanatory diagram illustrating an example of an image captured at the position (4) illustrated in FIG.
The notation method of the feature points is the same as in FIGS. 24A and 24B. The matching score is s = 1/8.

FIG. 28A is an explanatory diagram showing an image taken at the same position as FIG. 24A.
FIG. 28B is an explanatory diagram illustrating an example of an image captured at the position (5) illustrated in FIG.
The notation method of the feature points is the same as in FIGS. 24A and 24B. The matching score is s = 1/8.
Therefore, the target position determination unit 11d determines that the position in FIG. 25B that gives the maximum score is the target position.
FIG. 29 is an explanatory diagram visually showing how the feature points shown in FIG. 24A are extracted from the image.
The feature point extraction unit 11c binarizes the image captured by the imaging unit 133 and extracts the position of the pixel representing the corner as a feature point.

(Embodiment 2)
In the first embodiment, the target position determination unit 11d acquires the room coordinates of the candidate position by performing a nearest neighbor search of the position information database when the user learns the candidate position. As a different aspect, in the candidate position learning stage, feature point information and distance information are created from an image captured at that position, and the feature point information and distance information are associated with a keyword for identifying the candidate position and a storage unit 61 may be stored.
Thereafter, when receiving an instruction “clean the front of the TV” from the user, the candidate positions stored in the storage unit 61 are searched for candidate positions corresponding to the keyword “in front of the TV”, and the distance Get information. Then, the nearest neighbor search may be performed to find the position information having the most similar distance information in the position information database, and the room coordinates related to the position information may be obtained.
According to this aspect, the candidate position can be learned even if the position information database has not yet been created. The position information database only needs to be created by the next point at which an instruction to move to the learned candidate position is received.

(Embodiment 3)
In addition to creating the position information database in response to an instruction from the user, for example, when the self-propelled cleaner 1 is cleaning a room, the position information database may be created together. Good. If the location information database is checked and updated every time it is cleaned, it can be detected that the room has been redesigned. In this case, the user can be informed that the candidate position has to be redone because the room scene has changed.

(Embodiment 4)
Alternatively, feature point information and distance information may be stored in place of the room coordinates of the candidate position only when the candidate position is learned at the stage where the position information database has not yet been created. Then, when the position information database is completed, a near neighbor search may be performed to obtain the room coordinates of the candidate position, and the feature point information and the distance information may be replaced with the room coordinates.

(Embodiment 5)
In the above embodiment, the case of three reference points has been described, but the number of reference points is not necessarily limited to three. The number may be 3 or more. For example, when four reference points are used, a contiguous rectangle is searched. If there are more sides than triangles, the chances of finding congruents by chance are lessened, so accurate matching can be realized. However, since the amount of calculation of the distance increases when the number of sides is large, the designer should select an appropriate number.
When there are four reference points, the matching reference points may be three. Two points may be used, but in that case, the score is calculated for any combination of two other points, so the processing amount for calculating the score increases, but the score is calculated based on many combinations. Therefore, the accuracy of matching is high.
In the case of three reference points, the number of matching reference points is not limited to two and may be one. However, the amount of calculation of the score increases.

(Embodiment 6)
In the above embodiment, the correspondence between feature points is determined based on the distance between two points. This is based on the idea that the distance between two points is invariant to the Euclidean transformation. However, the present invention is not limited to the distance between two points, but can be applied to any invariant (Euclidean invariant) for Euclidean transformation. For example, since the angle between two straight lines and the area of the closed region are Euclidean invariants, any one of the invariants may be used instead of the distance between two points.
In addition, invariants under similarity transformations (similar invariants) and invariants under affine transformations (similar invariants), which can be used for similarity transformations, affine transformations and projective transformations, which are more flexible than Euclidean transformations. Affine invariants) or invariants (projective invariants) under projective transformation may be used.

As similar invariants, in addition to the angle between two straight lines, a distance ratio, an area ratio, and the like are known. As an affine invariant, the ratio of the distances of three points on the position line is known. A cross ratio is known as a projection invariant.
Using invariants for geometric transformation with a high degree of freedom makes it less susceptible to distortion of images taken at different positions (viewpoints), but increases the load of invariant calculation processing. The designer should appropriately select whether or not to apply matching.

As mentioned above,
(I) A self-propelled electronic device according to the present invention is photographed by a traveling control unit that controls traveling, a position detecting unit that detects a position that changes along with the traveling, and an imaging unit that captures a surrounding scene. A feature point extracting unit that extracts a plurality of feature points from the captured image, a feature point related to the image captured by the imaging unit at a plurality of positions in the traveling region, and a position where the image was captured are associated and registered A location information database storage unit that stores the location information in advance, a query storage unit that stores feature points related to an image captured by the imaging unit at a target position, and a feature point stored in the query storage unit A target that collates an arrangement relationship with feature points included in each position information stored in the position information database storage unit and determines a target position based on position information that includes the most feature points in the corresponding arrangement relationship A position determination unit, and the travel control unit controls travel to the target position based on the current position detected by the position detection unit and the determined target position.

In the present invention, the position detector detects a position that changes as the vehicle travels. The specific aspect is an encoder arrange | positioned at the drive motor which drives a wheel or a wheel, for example. When the drive motor is a stepping motor, drive pulses may be counted. Since the rotation direction of the wheel is reversed between the forward movement and the backward movement, a means capable of detecting the rotation direction is necessary.
The feature points are a plurality of points that represent the features of the image. Known techniques for extracting feature points from an image include SIFT, SURF, ORB, AKAZE, and the like, and these known algorithms can be applied.

  Furthermore, the feature point arrangement relationship means whether or not there is a corresponding relative positional relationship between feature points among a plurality of feature points of both images to be collated. To do. Here, “corresponding” may be within an allowable range even if some error is included. As a specific mode for evaluating the arrangement relationship of feature points, for example, there is a method of calculating whether or not the distance between two points is equal for all combinations with respect to a predetermined number of feature points. Not limited to.

Furthermore, the preferable aspect of this invention is demonstrated.
(Ii) The imaging unit sequentially captures images while traveling, and the traveling control unit applies the feature points relating to the images captured by the imaging unit during traveling and the images of the respective positions stored as the position information database. You may correct | amend the position which the said position detection part detects by collating with the feature point which concerns.
In this way, the current position is estimated by collating the feature point related to the image captured during traveling with the feature point related to the position information database, and the position detection unit uses the estimated current position. The position information to be detected can be corrected and the current position can be determined stably.

  (Iii) A candidate position storage unit that stores feature points related to images captured at one or more candidate positions in association with selection instructions used when selecting each candidate position as a target, and a travel to be traveled to the target An instruction receiving unit that receives the instruction, the travel instruction includes the selection instruction related to any candidate position, and the target position determination unit includes the selection instruction included in the travel instruction and the candidate position storage unit And the feature point associated with the corresponding selection instruction is stored in the query storage unit, and the travel control unit travels to the target selected in response to the travel instruction. May start.

In this way, by storing one or more candidate positions in the candidate position storage unit in advance, and then giving a traveling instruction including a selection instruction for selecting one of those candidate positions, The self-propelled electronic device can travel to the selected position.
For example, if the travel instruction is an instruction by voice, a selection instruction by unique voice such as “in front of the sofa”, “in front of the television”, “under the desk”, or the like is associated with the candidate position. Including any selection instruction among them, it is assumed that the voice “clean up” indicates that it is a driving instruction. Then, the self-propelled vacuum cleaner is a candidate position where the target position is related to “in front of the sofa” and the user's call “clean the front of the sofa”. Can be recognized. Therefore, it is possible to run in front of the sofa and work in the spot mode.

(Iv) The target position determination unit determines whether there is a combination of feature points having a corresponding arrangement relationship between the feature points stored in the query storage unit and the feature points included in each position information. If there is at least one combination of feature points in the corresponding arrangement relationship, a predetermined number of feature points related to the combination is used as a reference feature point, and at least one reference feature point and other features The target position may be determined on the basis of position information including the largest number of pairs of feature points in the corresponding arrangement relationship by examining the arrangement relationship of the feature points with respect to the combination of feature points including points.
In this way, when the feature points stored in the query storage unit and the feature points stored as the position information database are collated, the corresponding pairs of the feature points having the corresponding arrangement relationship exist. By examining how many, it is possible to determine the position related to the position information including the most pairs as the target position.

(V) The corresponding arrangement relationship is the length of all sides of a triangle formed by three points of three feature points stored in the query storage unit and three of the feature points included in each position information. Are determined based on whether or not the two are equal, and the target position determination unit uses three feature points forming the vertices of a pair of triangles corresponding to the arrangement relationship as reference feature points, and sets three reference feature points of each pair. For the triangle formed by two reference feature points and one feature point other than the reference feature points, the number of pairs corresponding to the arrangement relationship is further examined, and based on the position information including the most corresponding pairs. The target position may be determined.
In this way, it is possible to determine whether or not the feature point arrangement relationship corresponds by comparing the lengths of the sides of the triangle formed by the three feature points.
Preferred embodiments of the present invention include combinations of any of the plurality of embodiments described above.
In addition to the embodiments described above, there can be various modifications of the present invention. These modifications should not be construed as not belonging to the scope of the present invention. The present invention should include the meaning equivalent to the scope of the claims and all modifications within the scope.

1: Self-propelled cleaner, 2: Housing, 2b ₁ : Top plate front, 2b ₁₁ : Air hole, 2b ₂ : Lid, 2c ₁ : Front plate front half, 2c ₂ : Side plate rear half, 2d: Internal structure Wall: 9: Rotating brush 10: Side brush 11: Equipment control unit 11a: Travel control unit 11b: Position detection unit 11c: Feature point extraction unit 11d: Target position determination unit 11S: Circuit board 12 : Rechargeable battery, 13: Connecting part for charging, 14: Fault detecting part, 14A: Ultrasonic receiving part, 14B: Ultrasonic transmitting part, 14C: Collision sensor, 15: Dust collecting part, 15a: Dust collecting container, 15b: Dust collection filter, 18: floor detection sensor, 20: gyro sensor, 21L: left wheel drive motor, 21R: right wheel drive motor, 22L: left drive wheel, 22R: right drive wheel, 23: drive unit, 24: induction signal Receiver 26: Rear wheel 31: Suction port 32: First exhaust port 33: Second exhaust port 55: Notification unit 61: Storage unit 61a: Position information database storage unit 61b: Query storage unit 71L: Left wheel encoder 71R : Right wheel encoder, 114: Duct section, 115: Electric blower, 119: Brush motor, 120: Ion generator, 121L: Left wheel driver, 121R: Right wheel driver, 123: Blower driver 125: Brush driver, 131: Wireless communication 133: Imaging unit 135: Voice recognition unit 201: Charging stand 202: Charging terminal unit 203: Induction signal sending unit 300: Room 301: Table 303: Sofa 305: Television 307: Bookshelf 309 :Air cleaner

Claims (5)

A travel control unit for controlling travel;
A position detector that detects a position that changes with the travel;
An imaging unit that captures the surrounding scene;
A feature point extraction unit that extracts a plurality of feature points from the captured image;
A position information database storage unit that stores in advance position information registered in association with feature points related to images captured by the image capturing unit at a plurality of positions in the traveling region and positions where the image capturing was performed;
A query storage unit that stores feature points related to an image captured by the imaging unit at a target position;
A position that includes the largest number of feature points in the corresponding arrangement relation by collating the arrangement relation between the feature points stored in the query storage section and the feature points included in the position information stored in the position information database storage section. A target position determination unit that determines a target position based on information,
The traveling control unit is a self-propelled electronic device that controls traveling to the target position based on the current position detected by the position detection unit and the determined target position. The imaging unit performs sequential shooting while traveling,
The travel control unit collates a feature point related to an image captured by the imaging unit during travel with a feature point related to an image at each position stored as the position information database, and the position detection unit detects the feature point. The self-propelled electronic device according to claim 1 which corrects a position. A candidate position storage unit that stores feature points related to images taken at one or more candidate positions in association with selection instructions used when selecting each candidate position as a target;
An instruction receiving unit for receiving a driving instruction to drive to the target;
The travel instruction includes the selection instruction related to any candidate position,
The target position determination unit collates a selection instruction included in the travel instruction with a selection instruction stored in the candidate position storage unit, and stores a feature point associated with the corresponding selection instruction in the query storage unit. ,
The self-propelled electronic device according to claim 1, wherein the travel control unit starts traveling to a selected target in response to the travel instruction.   The target position determination unit examines whether or not there is a combination of feature points in a corresponding arrangement relationship between the feature points stored in the query storage unit and the feature points included in each position information, When there is at least one combination of feature points in the corresponding arrangement relationship, a predetermined number of feature points related to the combination is used as a reference feature point, and includes at least one reference feature point and other feature points 4. The target position is determined based on position information including the largest number of pairs of feature points in the corresponding arrangement relationship by examining the arrangement relationship of the feature points with respect to the combination of feature points. The self-propelled electronic device described in 1. Corresponding arrangement relation is that the length of all the sides of the triangle formed by the three points is the same for three of the feature points stored in the query storage unit and three of the feature points included in each position information. Based on whether or not
The target position determination unit uses three feature points that form the vertices of a pair of triangles corresponding to the arrangement relationship as reference feature points, and two of the three reference feature points of each pair other than the two reference feature points and the reference feature points 5. The target position is determined based on position information including the largest number of corresponding pairs, by examining how many pairs of the triangles formed by one feature point correspond to the arrangement relationship. Self-propelled electronic equipment.