KR20240146350A - Method for drawing up a map and apparatus thereof - Google Patents

Abstract

The present invention relates to a map-making method performed by a computing device, comprising: a step of collecting primary pose data; a step of extracting feature information of the collected primary pose data; a step of detecting one or more representative pose data based on the extracted feature information; and a step of collecting secondary pose data based on location information of the detected representative pose data.

Description

METHOD FOR DRAWING UP A MAP AND APPARATUS THEREOF

The present invention relates to a method for creating a map using a robot, and more specifically, to a method for searching for an optimal collection location of pose data required for creating a map of a robot.

Recently, as interest in robots capable of autonomous driving has increased, various robot products are being developed, and autonomous driving is a key keyword of the 4th industrial revolution, and it is a core technology with a wide range of applications, such as autonomous cars, delivery robots, and unmanned security robots. The core component of these autonomous robots is the navigation system, which allows the robot to detect the surrounding environment, efficiently avoid obstacles, and move along the shortest path.

The robot uses the SLAM (Simultaneous Localization and Mapping) technique, a representative autonomous driving algorithm, to create a map of the target space. The SLAM technique is a technique in which the robot recognizes the surrounding spatial terrain or artificial landmarks by utilizing various sensors installed on the robot as it moves, and uses the obtained spatial terrain or artificial landmarks to create a map of the surrounding environment while simultaneously determining the relative position of the robot.

As the robot moves through the target space, it collects pose data, including IMU (Inertial Measurement Unit) data, LiDAR data, and imaging data, at regular intervals or at regular distances. The pose data plays an important role in increasing the accuracy of map creation.

However, the map created by the robot may have errors with the actual map of the target space. Such errors may cause significant problems in the robot's position recognition and map creation. In particular, there is a possibility of false positives and false negatives depending on how the pose data collection interval is configured. For example, as shown in Fig. 1 (a), if the pose data collection interval is set densely, the possibility of false positives may increase because nearby points may be judged to be the same as the current point. On the contrary, as shown in Fig. 1 (b), if the pose data collection interval is set sparsely, the possibility of false negatives may increase. Therefore, pose data was collected by setting an optimal interval through a parameter tuning process by an expert in the past. However, in actual fields, the number of experts is limited, and there is a problem that the parameter tuning process takes a considerable amount of time.

KR 10-2021-0003065 A

The present invention aims to solve the above-mentioned problems and other problems. Another object is to provide a method and device for searching for an optimal collection location of pose data required for creating a map of a robot.

Another purpose is to provide a method and device for extracting feature information of a target space and searching for an optimal collection location of pose data based on the extracted feature information.

According to one aspect of the present invention to achieve the above or other purposes, a map creation method is provided, including: a step of collecting primary pose data; a step of extracting feature information of the collected primary pose data; a step of detecting one or more representative pose data based on the extracted feature information; and a step of collecting secondary pose data based on location information of the detected representative pose data.

According to another aspect of the present invention, a map creation device is provided, including: a data collection unit that collects pose data required for creating a map of a target space; a feature information extraction unit that extracts feature information of the collected pose data; a density estimation unit that calculates a probability density of the extracted feature information; and a data collection location detection unit that detects one or more representative pose data based on the calculated probability density.

According to another aspect of the present invention, a computer program is provided, which is stored in a computer-readable recording medium, so that the steps of: collecting primary pose data; extracting feature information of the collected primary pose data; detecting one or more representative pose data based on the extracted feature information; and collecting secondary pose data based on location information of the detected representative pose data can be executed on a computer.

The effects of a map creation method and its device according to embodiments of the present invention are described as follows.

According to at least one of the embodiments of the present invention, there is an advantage in that feature information of a target space can be extracted using lidar data of the target space, and an optimal collection location of pose data required for map creation can be detected based on the extracted feature information.

In addition, according to at least one of the embodiments of the present invention, there is an advantage in that feature information of a target space can be extracted using lidar data and imaging data of the target space, and an optimal collection location of pose data required for map creation can be detected based on the extracted feature information.

In addition, according to at least one of the embodiments of the present invention, there is an advantage in that the accuracy of map creation can be improved compared to the prior art by collecting pose data based on points where a lot of feature information of the target space exists.

However, the effects that can be achieved by the map-making method and the device according to the embodiments of the present invention are not limited to those mentioned above, and other effects that are not mentioned will be clearly understood by those skilled in the art to which the present invention belongs from the description below.

Figure 1 is a drawing referenced to explain the influence according to the collection interval of pose data;
FIG. 2 is a flowchart illustrating a map creation method according to one embodiment of the present invention;
Figure 3 is a drawing referenced to explain the map creation method of Figure 2;
FIG. 4 is a block diagram of a pose data collection position search device according to an embodiment of the present invention;
FIGS. 5 and 6 are drawings referenced to explain a method for extracting feature information of lidar data;
Figure 7 is a drawing referenced to explain a method for estimating the probability density of feature information;
Figure 8 is a drawing referenced to explain a method for detecting a collection location of pose data based on the probability density of feature information.
FIG. 9 is a flowchart explaining a method for searching a pose data collection location according to one embodiment of the present invention;
FIG. 10 is a block diagram of a pose data collection position search device according to another embodiment of the present invention;
FIG. 11 is a drawing referenced to explain a method for extracting feature information of lidar and imaging data;
Figure 12 is a drawing referenced to explain a method for estimating the probability density of feature information;
FIG. 13 is a drawing referenced to explain a method for detecting a collection location of pose data based on the probability density of feature information;
FIG. 14 is a flowchart explaining a method for searching a pose data collection location according to another embodiment of the present invention;
Figure 15 is a block diagram of a computing device according to an embodiment of the present invention.

Hereinafter, embodiments disclosed in the present specification will be described in detail with reference to the attached drawings. Regardless of the drawing symbols, identical or similar components will be given the same reference numerals and redundant descriptions thereof will be omitted. The suffixes "module" and "part" used for components in the following description are given or used interchangeably only for the convenience of writing the specification, and do not have distinct meanings or roles in themselves. That is, the term "part" used in the present invention means a hardware component such as software, FPGA, or ASIC, and the "part" performs certain roles. However, the "part" is not limited to software or hardware. The "part" may be configured to be in an addressable storage medium or may be configured to reproduce one or more processors. Thus, as an example, a 'part' may include components such as software components, object-oriented software components, class components and task components, processes, functions, attributes, procedures, subroutines, segments of program code, drivers, firmware, microcode, circuitry, data, databases, data structures, tables, arrays and variables. The functionality provided within the components and 'parts' may be combined into a smaller number of components and 'parts' or further separated into additional components and 'parts'.

In addition, when describing the embodiments disclosed in this specification, if it is determined that a detailed description of a related known technology may obscure the gist of the embodiments disclosed in this specification, the detailed description thereof will be omitted. In addition, the attached drawings are only intended to facilitate easy understanding of the embodiments disclosed in this specification, and the technical ideas disclosed in this specification are not limited by the attached drawings, and should be understood to include all modifications, equivalents, and substitutes included in the spirit and technical scope of the present invention.

The present invention proposes a method and a device for searching for an optimal collection location of pose data required for creating a map of a robot. In addition, the present invention proposes a method and a device for extracting feature information of a target space and searching for an optimal collection location of pose data based on the extracted feature information.

Hereinafter, various embodiments of the present invention will be described in detail with reference to the drawings.

FIG. 2 is a flowchart explaining a map creation method according to an embodiment of the present invention, and FIG. 3 is a drawing referred to for explaining the map creation method of FIG. 2. The map creation method according to the present embodiment can be performed by a map creation device (not shown). The map creation device can be installed in a robot or installed in an external server. In the illustrated flowchart, the map creation method is described by dividing it into a plurality of steps, but at least some of the steps may be performed in a changed order, combined with other steps and performed together, omitted, divided into detailed steps and performed, or one or more steps not shown may be added and performed.

Referring to FIG. 2, a map creation device according to an embodiment of the present invention can set a collection interval of pose data required for map creation (S210). The collection interval of the pose data can be set to a certain time cycle or a certain distance unit.

The map-making device can primarily collect pose data according to a preset collection interval while moving through the target space (10) (S220). The pose data to be collected may include lidar data, imaging data, and IMU (Inertial Measurement Unit) data measured at a specific point in the target space.

When the robot has completed moving the entire target space (10), the map creation device can detect the optimal collection location of pose data based on the primarily collected pose data (S230). A detailed description of the method for searching the optimal collection location of the pose data will be described later.

Thereafter, the map creation device can secondarily collect pose data at the detected optimal collection location while moving again through the target space (10) (S240).

The map creation device can create a map based on secondarily collected pose data (S250).

The present invention collects pose data for a target space not at a set time interval or a set distance unit, but based on points where there is a lot of feature information (or feature data) of the target space, thereby improving the accuracy of map creation and alignment (LoopClosure) compared to the prior art.

Figure 4 is a block diagram of a pose data collection location search device according to an embodiment of the present invention. The pose data collection location search device can be implemented within a map creation device.

Referring to FIG. 4, a pose data collection location search device (100) according to an embodiment of the present invention may include a data collection unit (110), a feature information extraction unit (120), a density estimation unit (130), and a data collection location detection unit (140). The components illustrated in FIG. 4 are not essential for implementing the pose data collection location search device, and thus, the pose data collection location search device described in this specification may have more or fewer components than the components listed above.

The data collection unit (110) can collect LiDAR data for the target space through a LiDAR sensor mounted on the robot. The LiDAR data can be collected at a certain time period or at a certain distance interval.

The feature information extraction unit (120) can convert the lidar data into two-dimensional image data and extract feature information of the converted image data. At this time, the feature information extraction unit (120) can extract feature information of the converted image data using a pre-learned feature inference model (or dimension reduction model). As the feature inference model, a PCA (principal component analysis) model, an autoencoder model, a GAN (Generative Adversarial Network) model, etc. can be used, and more preferably, an autoencoder model can be used.

For example, as illustrated in FIG. 5, the feature information extraction unit (120) can convert the lidar data (510) into grid-shaped image data (520). The feature information extraction unit (120) can divide the grid-shaped image data (520) into four image areas (531 to 534).

The feature information extraction unit (120) can input the first to fourth segmented image areas (531 to 534) into the first to fourth feature inference models (541 to 544), respectively, to extract the first to fourth segmented feature information (551 to 554). The feature information extraction unit (120) can connect (concatenate) the extracted first to fourth segmented feature information (551 to 554) to generate one feature information (560), i.e., feature information of lidar data.

The first to fourth feature inference models (541 to 544) can be implemented through an autoencoder algorithm. The autoencoder algorithm is an unsupervised learning technique that converts inputs into signals through an encoder and then creates labels and the like through a decoder. For example, as shown in (a) of Fig. 6, lidar image data (610) converted from lidar data into two-dimensional images and restored image data (620) can be collected, and the collected data can be learned to generate an autoencoder model (630).

In the present invention, the encoder part (635) of the autoencoder model (630) can be used as a feature inference model. That is, as shown in (b) of Fig. 6, the lidar image data (640) can be input into the pre-learned feature inference model (635) to extract feature information (650) of the corresponding image data.

The density estimation unit (130) estimates the probability density of feature information corresponding to the lidar data using a predetermined density estimation algorithm, and can compare the feature information of the lidar data with each other based on the estimated probability density information. Here, a Gaussian kernel density estimation algorithm can be used as the density estimation algorithm, but is not necessarily limited thereto.

For example, as illustrated in FIG. 7, the density estimation unit (130) can calculate the probability density of feature information (711 to 715) for the lidar data using a Gaussian kernel density estimation algorithm. The calculated probability density can be displayed in a predetermined graph form. That is, the probability density of the first to fifth feature information (711 to 715) can be displayed as the first to fifth graphs (721 to 725).

The density estimation unit (130) can set one or more reference feature information from among multiple feature information based on probability density information. For example, the feature information (711) input first is set as the reference feature information, and if there is feature information that has a difference in probability density greater than a threshold value compared to the previous reference feature information, the corresponding feature information (713) can be set as new reference feature information.

The density estimation unit (130) can compare each reference feature information with its adjacent feature information based on the probability density information. At this time, the density estimation unit (130) can measure the similarity between the probability density of the reference feature information and the probability density of its adjacent feature information, and compare the reference feature information with its adjacent feature information based on the measured similarity.

The density estimation unit (130) can compare feature information corresponding to lidar data based on probability density information and classify the feature information into multiple groups. At this time, the density estimation unit (130) can classify the feature information based on similarity information between the feature information. Accordingly, feature information having the same or similar probability density pattern can be grouped together.

Each group may include reference feature information and adjacent feature information that is identical or similar to the reference feature information. The number of groups corresponds to the number of reference feature information and represents the complexity of the map.

For example, as illustrated in FIG. 7, the density estimation unit (130) can generate a first group (731) by comparing the probability density (721) of the first reference feature information (711) with the probability densities of its adjacent feature information. In addition, the density estimation unit (130) can generate a second group (732) by comparing the probability density (723) of the second reference feature information (713) with the probability densities of its adjacent feature information.

The data collection location detection unit (140) can detect representative lidar data (i.e., representative pose data) based on multiple group information. At this time, the data collection location detection unit (140) can detect representative pose data for each group.

For example, as illustrated in FIG. 8, the data collection location detection unit (140) can detect representative feature information of each group (731, 732) and detect representative pose data (741, 742) corresponding to the detected representative feature information.

The data collection location detection unit (140) can detect location information where representative pose data is collected. Similarly, the data collection location detection unit (140) can detect location information of representative pose data for each group.

As described above, the pose data collection location search device according to one embodiment of the present invention can extract feature information of a target space using lidar data of the target space, and detect an optimal collection location of pose data required for map creation based on the extracted feature information.

FIG. 9 is a flowchart explaining a pose data collection location search method according to an embodiment of the present invention. The pose data collection location search method according to the present embodiment can be performed by a pose data collection location search device (100). In the illustrated flowchart, the pose data collection location search method is described by dividing it into a plurality of steps, but at least some of the steps may be performed in a changed order, combined with other steps and performed together, omitted, divided into detailed steps and performed, or one or more steps not illustrated may be added and performed.

Referring to FIG. 9, a pose data collection location search device (100) according to an embodiment of the present invention can collect LiDAR data for a target space through a LiDAR sensor mounted on a robot (S910). The LiDAR data can be collected at a certain time cycle or a certain distance interval.

The pose data collection position search device (100) can convert the lidar data into two-dimensional image data (S920). For example, the pose data collection position search device (100) can convert the lidar data into grid-shaped image data.

The pose data collection location search device (100) can extract feature information of the converted image data using a pre-learned feature inference model (S930). The feature inference model may include a PCA model, an autoencoder model, a GAN model, etc., and more preferably, an autoencoder model may be used.

The pose data collection location search device (100) can calculate the probability density of feature information corresponding to lidar data using a predetermined density estimation algorithm (S940). A Gaussian kernel density estimation algorithm can be used as the density estimation algorithm, but is not necessarily limited thereto.

The pose data collection location search device (100) can compare feature information corresponding to lidar data with each other based on probability density information and classify the feature information into multiple groups (S950). At this time, the pose data collection location search device (100) can compare each reference feature information with its adjacent feature information based on probability density information.

The pose data collection location search device (100) can detect representative lidar data (i.e., representative pose data) based on multiple group information (S960). At this time, the pose data collection location search device (100) can detect representative pose data for each group.

The pose data collection location search device (100) can detect location information where representative pose data is collected (S970). Similarly, the pose data collection location search device (100) can detect location information of representative pose data for each group.

As described above, the pose data collection location search method according to one embodiment of the present invention can extract feature information of a target space using lidar data of the target space, and detect the optimal collection location of pose data required for map creation based on the extracted feature information.

Fig. 10 is a block diagram of a pose data collection location search device according to another embodiment of the present invention. The pose data collection location search device can be implemented within a map creation device.

Referring to FIG. 10, a pose data collection location search device (200) according to another embodiment of the present invention may include a data collection unit (210), a feature information extraction unit (220), a density estimation unit (230), and a data collection location detection unit (240). The components illustrated in FIG. 10 are not essential for implementing the pose data collection location search device, and thus, the pose data collection location search device described in this specification may have more or fewer components than the components listed above.

The data collection unit (210), feature information extraction unit (220), density estimation unit (230), and data collection location detection unit (240) are similar to the data collection unit (110), feature information extraction unit (120), density estimation unit (130), and data collection location detection unit (140) of FIG. 4 described above, so detailed descriptions of the data collection unit (210), feature information extraction unit (220), density estimation unit (230), and data collection location detection unit (240) will be omitted and explanations will be focused on the differences between them.

The data collection unit (210) may include a lidar data collection unit (211) that collects lidar data and an imaging data collection unit (212) that collects imaging data.

The LiDAR data collection unit (211) can collect LiDAR data for a target space through one or more LiDAR sensors mounted on the robot. The LiDAR data can be collected at a certain time period or a certain distance interval.

The imaging data collection unit (212) can collect imaging data for a target space through one or more cameras mounted on the robot. The imaging data can also be collected at a certain time period or at a certain distance interval.

The feature information extraction unit (220) may include a lidar data feature extraction unit (221) that extracts feature information of lidar data, and an imaging data feature extraction unit (222) that extracts feature information of imaging data.

The lidar data feature extraction unit (221) can convert lidar data into two-dimensional image data and extract feature information of the converted image data. At this time, the lidar data feature extraction unit (221) can extract feature information of the converted image data using a pre-learned feature inference model. As the feature inference model, a PCA model, an autoencoder model, a GAN model, etc. can be used, and more preferably, an autoencoder model can be used.

For example, as illustrated in FIG. 11, the lidar data feature extraction unit (221) can convert the lidar data (1111, 1112) into grid-shaped image data (1121, 1122). The lidar data feature extraction unit (221) can extract feature information (1131, 1132) of the grid-shaped image data (1121, 1122) using a pre-learned autoencoder model. Here, the feature information can be a 25-dimensional feature vector, but is not necessarily limited thereto. That is, a typical autoencoder model can extract features of the input grid-shaped image data (1121, 1122) and output a 25-dimensional feature vector.

Meanwhile, although not shown in the drawing, the lidar data feature extraction unit (221) can divide each grid-shaped image data (1121, 1122) into multiple image areas, extract feature information of the divided image areas, and then combine the feature information.

The imaging data feature extraction unit (222) can extract feature information of imaging data using a pre-learned feature inference model. As the feature inference model, a SIFT (Scale-Invariant Feature Transform) model, a SURF (Speeded Up Robust Features) model, etc. can be used, and more preferably, a SIFT model can be used.

For example, as illustrated in FIG. 11, the imaging data feature extraction unit (222) can extract feature information (1151, 1152) of imaging data (1141, 1142) using a pre-learned SIFT model. Here, the feature information may be a 128-dimensional feature vector, but is not necessarily limited thereto. That is, a typical SIFT model can extract features of input imaging data (1141, 1142) and output a 128-dimensional feature vector.

The density estimation unit (230) can generate a single integrated feature information by combining the feature information of the lidar data and the feature information of the imaging data. In one embodiment, the density estimation unit (230) can generate a 153-dimensional integrated feature vector by concatenating a 25-dimensional feature vector and a 128-dimensional feature vector. Meanwhile, in another embodiment, when the grid-shaped image data is divided into four regions and an autoencoder model is applied to each divided region, the density estimation unit (230) can also generate a 228-dimensional integrated feature vector by concatenating a 100-dimensional (25*4)-dimensional feature vector and a 128-dimensional feature vector.

The density estimation unit (230) calculates the probability density of integrated feature information corresponding to the lidar and imaging data using a predetermined density estimation algorithm, and compares the integrated feature information of the lidar and imaging data based on the calculated probability density information. Here, a Gaussian kernel density estimation algorithm may be used as the density estimation algorithm, but is not necessarily limited thereto.

For example, as illustrated in FIG. 12, the density estimation unit (230) can calculate the probability density of the integrated feature information (1211 to 1214) for the lidar and imaging data using a Gaussian kernel density estimation algorithm. The calculated probability density can be displayed in a predetermined graph form. That is, the probability density of the first to fourth integrated feature information (1211 to 1214) can be displayed as the first to fourth graphs (1221 to 1224).

The density estimation unit (230) can set one or more reference integrated feature information from among multiple pieces of integrated feature information based on probability density information. For example, the first input integrated feature information (1211) can be set as reference feature information, and if there is integrated feature information that has a probability density difference greater than a threshold value compared to the previous reference integrated feature information, the corresponding integrated feature information (1213) can be set as new reference integrated feature information.

The density estimation unit (230) can compare the integrated feature information corresponding to the lidar and imaging data based on the probability density information and classify the integrated feature information into multiple groups. At this time, the density estimation unit (230) can classify the integrated feature information into multiple groups based on the similarity information between the integrated feature information. Accordingly, feature information having the same or similar probability density pattern can be grouped together.

For example, as illustrated in FIG. 12, the density estimation unit (230) can generate a first group (1231) by comparing the probability density (1221) of the first reference integrated feature information (1211) with the probability densities of its adjacent integrated feature information. In addition, the density estimation unit (230) can generate a second group (1232) by comparing the probability density (1223) of the second reference integrated feature information (1213) with the probability densities of its adjacent feature information.

The data collection location detection unit (240) can detect representative pose data based on multiple group information. For example, as illustrated in FIG. 13, the data collection location detection unit (240) can detect representative integrated feature information of each group (1231, 1232) and detect representative pose data (1241, 1242) corresponding to the detected representative integrated feature information. At this time, the representative pose data can include representative lidar data and representative imaging data.

The data collection location detection unit (240) can detect location information where representative pose data is collected. At this time, the data collection location detection unit (240) can detect location information of representative pose data for each group.

As described above, a pose data collection location search device according to another embodiment of the present invention can extract feature information of a target space using lidar data and imaging data of the target space, and detect an optimal collection location of pose data required for map creation based on the extracted feature information.

FIG. 14 is a flowchart illustrating a pose data collection location search method according to another embodiment of the present invention. The pose data collection location search method according to the present embodiment can be performed by a pose data collection location search device (200). In the illustrated flowchart, the pose data collection location search method is described by dividing it into a plurality of steps, but at least some of the steps may be performed in a changed order, combined with other steps and performed together, omitted, divided into detailed steps and performed, or one or more steps not illustrated may be added and performed.

Referring to FIG. 14, a pose data collection position search device (200) according to another embodiment of the present invention can collect LiDAR data for a target space through one or more LiDAR sensors mounted on a robot (S1410). The LiDAR data can be collected at a certain time period or a certain distance interval.

The pose data collection location search device (200) can collect image data for a target space through one or more cameras mounted on the robot (S1420). The image data can also be collected at a certain time period or at a certain distance interval.

The pose data collection location search device (200) can convert lidar data into two-dimensional image data and extract feature information of the converted image data (S1430). At this time, the pose data collection location search device (200) can extract feature information of the converted image data using a pre-learned feature inference model. As the feature inference model, a PCA model, an autoencoder model, a GAN model, etc. can be used, and more preferably, an autoencoder model can be used.

The pose data collection location search device (200) can extract feature information of the captured data using a pre-learned feature inference model (S1440). The feature inference model may include a SIFT model, a SURF model, etc., and more preferably, a SIFT model may be used.

The pose data collection location search device (200) can generate a single integrated feature information by combining the feature information of lidar data and the feature information of imaging data (S1450).

The pose data collection location search device (200) can calculate the probability density of integrated feature information corresponding to lidar and imaging data using a predetermined density estimation algorithm (S1460). A Gaussian kernel density estimation algorithm can be used as the density estimation algorithm, but is not necessarily limited thereto.

The pose data collection location search device (200) can compare integrated feature information corresponding to lidar and imaging data based on probability density information and classify the feature information into multiple groups (S1470). At this time, the pose data collection location search device (200) can compare each reference integrated feature information with its adjacent integrated feature information based on probability density information.

The pose data collection location search device (200) can detect representative pose data based on multiple group information (S1480). At this time, the representative pose data can include representative lidar data and representative imaging data.

The pose data collection location search device (200) can detect location information where representative pose data is collected (S1490). At this time, the pose data collection location search device (200) can detect location information of representative pose data for each group.

As described above, a method for searching for a pose data collection location according to another embodiment of the present invention can extract feature information of a target space using lidar data and imaging data of the target space, and detect an optimal collection location of pose data required for map creation based on the extracted feature information.

FIG. 15 is a block diagram of a computing device according to an embodiment of the present invention.

Referring to FIG. 15, a computing device (1500) according to an embodiment of the present invention includes at least one processor (1510), a computer-readable storage medium (1520), and a communication bus (1530). The computing device (1500) may be one or more components included in the map creation device described above or the elements constituting the map creation device. In addition, the computing device (1500) may be one or more components included in the pose data collection location search device (100, 200) or the elements constituting the pose data collection location search device (100, 200).

The processor (1510) may cause the computing device (1500) to operate according to the exemplary embodiments described above. For example, the processor (1510) may execute one or more programs (1525) stored in a computer-readable storage medium (1520). The one or more programs may include one or more computer-executable instructions, which, when executed by the processor (1510), may be configured to cause the computing device (1500) to perform operations according to the exemplary embodiments.

A computer-readable storage medium (1520) is configured to store computer-executable instructions or program code, program data, and/or other suitable forms of information. A program (1525) stored in the computer-readable storage medium (1520) includes a set of instructions executable by the processor (1510). In one embodiment, the computer-readable storage medium (1520) may be a memory (e.g., volatile memory such as random access memory, non-volatile memory, or a suitable combination thereof), one or more magnetic disk storage devices, optical disk storage devices, flash memory devices, any other form of storage medium that can be accessed by the computing device (1500) and capable of storing desired information, or a suitable combination thereof.

A communication bus (1530) interconnects various other components of the computing device (1500), including the processor (1510) and computer-readable storage medium (1520).

The computing device (1500) may also include one or more input/output interfaces (1540) that provide interfaces for one or more input/output devices (1550) and one or more network communication interfaces (1560). The input/output interfaces (1540) and the network communication interfaces (1560) are connected to a communication bus (1530).

The input/output device (1550) can be connected to other components of the computing device (1500) via the input/output interface (1540). Exemplary input/output devices (1550) can include input devices such as a pointing device (such as a mouse or a trackpad), a keyboard, a touch input device (such as a touchpad or a touchscreen), a voice or sound input device, various types of sensor devices and/or photographing devices, and/or output devices such as a display device, a printer, a speaker, and/or a network card. The exemplary input/output device (1550) can be included within the computing device (1500) as a component that constitutes the computing device (1500), or can be connected to the computing device (1500) as a separate device distinct from the computing device (1500).

The above-described present invention can be implemented as a computer-readable code on a medium in which a program is recorded. The computer-readable medium may be one that continuously stores a computer-executable program or one that temporarily stores it for execution or downloading. In addition, the medium may be various recording means or storage means in the form of a single or multiple hardware combinations, and is not limited to a medium directly connected to a computer system, and may be distributed on a network. Examples of the medium may include magnetic media such as hard disks, floppy disks, and magnetic tapes, optical recording media such as CD-ROMs and DVDs, magneto-optical media such as floptical disks, and ROMs, RAMs, flash memories, etc., configured to store program instructions. In addition, examples of other media may include recording media or storage media managed by app stores that distribute applications or other sites, servers, etc. that supply or distribute various software. Therefore, the above detailed description should not be construed as limiting in all respects and should be considered exemplary. The scope of the present invention should be determined by a reasonable interpretation of the appended claims, and all changes coming within the equivalent scope of the present invention are intended to be included within the scope of the present invention.

100/200: Pose data collection location search device 110/210: Data collection unit
120/220: Feature information extraction section 130/230: Density estimation section
140/240: Data collection location detection unit 1500: Computing device

Claims (18)

In a method of creating a map performed by a computing device,
Step 1: Collecting pose data;
A step of extracting feature information of the collected primary pose data;
A step of detecting one or more representative pose data based on the extracted feature information; and
A map creation method comprising a step of collecting secondary pose data based on location information of the above-detected representative pose data. In the first paragraph, the first pose data collection step is,
A map creation method characterized by collecting the above first pose data at regular time intervals or at regular intervals. In the first paragraph,
A map creation method, characterized in that the first and second pose data include LiDAR data. In the third paragraph, the feature information extraction step,
A step of converting the above lidar data into two-dimensional image data; and
A map creation method characterized by including a step of extracting feature information of the converted image data using a pre-learned feature inference model. In paragraph 4,
A map creation method characterized in that the above feature inference model is implemented through an autoencoder algorithm. In the third paragraph, the representative pose data detection step,
A map creation method further comprising a step of calculating a probability density for feature information of the lidar data using a predetermined density estimation algorithm. In Article 6,
A map creation method characterized in that the above density estimation algorithm is a Gaussian kernel density estimation algorithm. In the sixth paragraph, the representative pose data detection step,
A step of classifying a plurality of feature information corresponding to a plurality of lidar data into one or more groups based on the above-described probability density; and
A map creation method characterized by comprising a step of detecting representative lidar data based on the classified group information. In the 8th paragraph, the step of classifying into one or more groups comprises:
A step of setting one or more reference feature information among the above multiple feature information; and
A map creation method characterized by including a step of classifying the plurality of feature information into one or more groups by comparing the probability density of the above-mentioned reference feature information with the probability density of its adjacent feature information. In the third paragraph,
A map creation method, characterized in that the first and second pose data include lidar data and imaging data. In the 10th paragraph, the feature information extraction step,
A step of extracting feature information of the lidar data using a pre-learned first feature inference model; and
A map creation method characterized by including a step of extracting feature information of the imaging data using a pre-learned second feature inference model. In Article 11,
The above first feature inference model is implemented through an autoencoder algorithm,
The above second feature inference model is a map creation method implemented through the SIFT (Scale-Invariant Feature Transform) algorithm. In Article 11,
A map creation method further comprising a step of generating integrated feature information by combining feature information of the extracted lidar data and feature information of the extracted imaging data. A computer program stored on a computer-readable recording medium so that the method according to any one of claims 1 to 13 can be performed on a computer. A data collection unit that collects pose data required for creating a map of the target space;
A feature information extraction unit that extracts feature information of the above collected pose data;
A density estimation unit that calculates the probability density of the extracted feature information; and
A map creation device including a data collection location detection unit that detects one or more representative pose data based on the probability density calculated above. In Article 15, the data collection unit,
A LiDAR data collection unit that collects LiDAR data through one or more LiDAR sensors; and
A map-making device characterized by including an imaging data collection unit that collects imaging data through one or more cameras. In Article 16, the specific information extraction unit,
A lidar data feature extraction unit for extracting feature information of the collected lidar data; and
A map creation device characterized by including an imaging data feature extraction unit that extracts feature information of the collected imaging data. In Article 15,
A map creation device characterized in that the data collection unit re-collects pose data based on location information of the detected representative pose data.

Images