JPH08204400A - Image recognition - Google Patents

Abstract

PURPOSE: To perform template matching having a short average recognition time by learning to align search areas in the order of the height of detecting frequency of a recognition object image on the basis of the past recognition results so as to replace an alignment of a search order with the learned order. CONSTITUTION: Positional recognition of an image of a recognition object 4 on the basis of the present search area and search order. Then, detection positions (x, y) of the registration points are recorded on an image recognition device body 1. Respective search areas are rearranged in the order of the higher number of detection from the distribution state to the respective search areas of the detection positions (x, y) of the registration points of the past fixed number so as to make a found order a prescribed research order. The search order having the shortest average recognition time required by learning replaces a conventional search order for being recorded on an image recognition device body.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image recognition method for recognizing a recognition target object and recognizing its position in mounting electronic parts, assembling parts, processing, etc., and more particularly to an image recognition method by template matching. Is.

[0002]

2. Description of the Related Art In recent years, due to high integration of ICs and miniaturization of IC patterns, it has become difficult to select a characteristic pattern within a recognition visual field by a recognition method based on binary template matching, resulting in erroneous recognition or recognition. There is a high probability that an error will occur. For this reason, template matching using grayscale images has been widely used in recent technologies.

However, in template matching of a grayscale image which requires processing of multi-tone data, the amount of calculation for matching tends to increase and the recognition time tends to be longer than in the case of binarization. In order to reduce the time, parallel processing of matching and pipeline processing are performed by increasing the hardware. There is also a method in which matching between the template and the recognition target image is not performed by comparing all pixels, but pixels are thinned out. Further, there is also a method of narrowing the search area to the position displacement range of the recognition target without searching the entire camera visual field.

[0004]

However, there is a problem that speeding up the recognition by increasing the above hardware increases the cost of the recognition device.

Further, the speeding up of recognition performed by thinning out matching pixels causes a problem of deterioration of recognition rate and recognition accuracy.

Further, speeding up recognition by narrowing down the search area has a problem that it becomes difficult to recognize a recognition object outside the search area, and it is often difficult to determine an optimum search area size.

The above-mentioned methods are properly used according to their characteristics, but any of them still has a problem.

The present invention solves the above-mentioned problems, reduces cost increase, does not deteriorate the recognition rate and recognition accuracy, does not leave the recognition object out of the search area, and achieves a short average recognition time by template matching. The subject is to provide an image recognition method.

[0009]

In order to solve the above-mentioned problems, the image recognition method of the first invention of the present application divides an image to be searched into a plurality of search areas in the image recognition method by template matching, A teaching step of teaching to search the search area in a predetermined search order,
A recognition step of searching an image position of a recognition target object based on the search area and a search order, a recording step of recording the recognition result, and detection of a recognition target image based on a past recognition result after a predetermined number of times of recognition. It is characterized by comprising a learning step of learning to arrange the search areas in descending order of frequency and a search order changing step of rearranging the previously taught search order in the learned order.

In order to solve the above-mentioned problems, the image recognition method of the second invention of the present application divides the image to be searched into a plurality of search areas in the image recognition method by template matching, and makes a predetermined search in each search area. A teaching step of teaching to search in order, a recognition step of searching an image position of a recognition object based on the search area and a search order, and a recording step of recording the recognition result,
After a predetermined number of recognitions, a first learning step of learning the previously taught search area so as to match the statistical distribution of past recognition target image detection positions, and after a predetermined number of recognitions, match the statistical distributions. The second learning step of learning so that the search areas set as described above are arranged in the descending order of the detection frequency of the recognition target images based on the past recognition results, the previously taught search area and the search order, And a search area / search order changing step of replacing the learned search area with the search order.

In order to solve the above-mentioned problems, the image recognition method of the third invention of the present application gradually increases or decreases the search area learned in the first learning step in the second invention of the present application. A third learning step for learning the size of the search area and the search order that minimize the average recognition time in the past recognition results is added.

[0012]

According to the image recognition method of the first invention of the present application, in the image recognition method by template matching, an instruction to divide an image to be searched into a plurality of search areas and to search each search area in a predetermined search order is taught. By the process, for example, when the recognition target object tends to be placed around the specific position, the search area and the search order can be taught according to the specific position, and the average recognition time can be shortened.

Then, after a predetermined number of times of recognition, a learning step of learning to arrange the search areas in the descending order of the detection frequency of the recognition target image based on the past recognition result and the search order taught previously are By the search order changing step of rearranging in the learned order, the average recognition time when using the search area taught at the beginning can be shortened.

Further, since the hardware is not particularly increased, the cost is low.

Further, since the search is performed until recognition, there is no omission of search, and there is no thinning of pixels, so that the recognition rate and recognition accuracy are not deteriorated.

The image recognition method of the second invention of the present application is the same as that of the first invention of the present application.
In addition to the action of the invention, a first learning step of learning the previously taught search area so as to match the statistical distribution of past recognition target image detection positions after a predetermined number of recognitions, and after a predetermined number of recognitions Second, learning is performed so that search areas set according to the statistical distribution are arranged in descending order of detection frequency of recognition target images based on past recognition results.
, And the search area / search order changing step of replacing the previously taught search area and search order with the learned search area and search order, for example, the recognition target object is located at a specific position. If there is a tendency to place it around the center, it is possible to rationally shorten the average recognition time while changing the size of the search area taught at the beginning according to the tendency of the position shift. it can.

The image recognition method of the third invention of the present application is the same as that of the second invention of the present application.
In addition to the operation of the invention, the search area learned in the first learning step is gradually increased or decreased to reduce the size of the search area and the search order in which the average recognition time is the shortest in the past recognition result. According to the third learning step of learning, for example, when the recognition target object tends to be placed around a specific position, the size of the search area taught at the beginning and the search order are set to the position of the position to be placed. The average recognition time can be shortened most reasonably while finely changing according to the tendency of deviation.

[0018]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS A first embodiment of the present invention will be described with reference to FIGS.

FIG. 1 is a block diagram showing the configuration of an image recognition device in an electronic circuit mounting facility using the method of the present invention. Reference numeral 1 is an image recognition device composed of a CPU, an image memory, an image signal input / output circuit, and the like. A main body, 2 is a camera, 3 is a monitor TV, 4 is a recognition target such as a board or component, 5 is a robot for moving the recognition target, 6 is a controller of the robot, and 7 is a control panel.

In the standard sample setting process of step # 1 of FIG. 2 showing each teaching process of the present embodiment, the reference sample of the recognition object 4 such as a substrate or a component is placed on the robot 5 and the operation panel 7 is operated. Then, the reference sample is set to the camera 2
, And proceed to step # 2.

In the alignment point teaching process of step # 2, a point serving as a reference for the position of the recognition object 4 displayed on the screen of the monitor television 3 (hereinafter referred to as alignment point).
By operating the graphics cross mark on the screen of the monitor TV 3 on the operation panel 7 and teaching it on Step # 3.
Proceed to.

In the search area / search order teaching step of step # 3, the operation panel 7 is operated to operate the monitor television 3.
Screen is taught to a plurality of search areas, for example, as shown in FIG. 5, divided into 9 areas and searched in a predetermined order (1) to (9), and the process proceeds to step # 4.

In FIG. 5, 8 is a monitor screen, 9 is a search area, and the numbers (1) to (9) in the figure indicate the search order of the search area.

In the template teaching process of step # 4, the graphics window on the screen of the monitor TV 3 is operated by the operation panel 7 to specify a specific small image of the recognition object 4 displayed on the screen of the monitor TV 3. The area is registered as template data in the image recognition apparatus body 1, and each teaching process is completed.

The recognition operation of this embodiment will be described with reference to FIG.

In the recognition process of step # 5 in FIG.
The position of the image of the recognition object 4 is recognized based on the active search area and the search order, and the process proceeds to step # 6.

In the recognition result recording step of step # 6, the detected position (x, y) of the alignment point is recorded in the image recognition apparatus main body 1, and the process proceeds to step # 7.

In the learning execution necessity determination step of step # 7, it is determined whether or not learning is to be executed. If the number of times of learning execution has been reached in advance, the process proceeds to step # 8 to learn, and it must be reached. If so, go to step # 10.

In the learning step of step # 8, each search area is rearranged in order from the one having the largest number of detections based on the distribution state of the detection positions (x, y) of the above-mentioned alignment points of a certain number of times in the past with respect to each search area. , The determined order of (1) to (9) is set as a predetermined search order, and step #
Proceed to 9.

In the search order changing step of step # 9, the search order, which has the shortest average recognition time obtained in the learning step, is registered in the image recognition apparatus main body by replacing the search order so far, with step # 10. move on.

In the process of judging whether the production is finished in step # 10, if the production is not finished, the process returns to step # 5, and if the production is finished, the process is finished.

A second embodiment of the present invention using the image recognition apparatus shown in FIG. 1 will be described with reference to FIGS.

1 and 2 are similar to those of the first embodiment, the description thereof will be omitted.

Also in FIG. 4, steps # 11 to # 13 are the same as steps # 5 to # 7 of the first embodiment, and therefore the description thereof will be omitted. The third and subsequent learning steps will be described.

In the first, second, and third learning steps of step # 14, in the first learning step, the average of the X coordinate and the Y coordinate of the detection positions (x, y) of the registration points of a certain number of times in the past is averaged. Values and standard deviations are obtained, and a first divided search area (1) whose size is determined so as to include the standard deviation (Sx, Sy) around the coordinate average value (Mx, My) as shown in FIG. Make a tentative decision. In FIG. 6, 8 is a monitor screen, 9 is a search area, and 10 is a detection position of the alignment point.

Next, in the second learning step, the second divided search area (2) to the ninth divided search area (9) are
As shown in FIG. 6, it is set at a position surrounding the first divided search area (1), and the search order is rearranged in descending order of the number of past position detection times in each divided search area.

Next, in the third learning step, when the divided search areas (1) to (9) set above are used, what is the average recognition time of the past fixed number of times of position recognition is determined. , Based on the following calculation formula (1) or (2).

[0038]

[Equation 1]

V _D : Average recognition time of divided search areas V _W : Recognition time of all search areas S _D (i): Area of divided search areas (i) _SW : Area of all search areas R _D (i): the number of detected R _W in the divided search area (i): the entire detection number NOTE division search area (i) is assumed to be aligned in order from those with many detection number.

The formula (1) is an approximate calculation formula, but the calculation time is short and it can be sufficiently used.

Expression (2) is a regular calculation expression. And
Using these formulas, the first divided search area (1)
Gradually increase or decrease the size of, and
The average recognition time is calculated when the search order of each search area is changed. Once the calculations are complete, step # 1
Go to 5. Regarding the above calculation formula, of course, a normal calculation formula may be used, but it is not always necessary, and an approximate calculation formula that can judge the trend is used to shorten the calculation time to reduce the productivity. Can be improved. Therefore,
The calculation formula is not limited to the one used in the embodiment, and may be designed according to the conditions.

In the search area and search order changing step of step # 15, the division size and search order of the search area, which is obtained in the learning step and which has the shortest average recognition time, are compared with the division size of the search area and the search. The order is changed to the registration order in the image recognition apparatus main body, and the process proceeds to step # 16.

In the step of judging whether the production is finished in step # 16, if the production is not finished, the process returns to step # 11, and if the production is finished, the process is finished.

In the embodiment, the search area is set to 9
Although it is divided, the number of divisions can be set freely. Further, when there are a plurality of recognition objects 4, it is preferable to set search areas separately.

[0045]

According to the image recognition method of the present invention, the field of view of the recognition camera is divided into a plurality of search areas, and learning is performed.
Until the position of the recognition target is detected, the search area is scanned in the descending order of the probability of the recognition target being detected, so that the position of the recognition target can be detected on average in the shortest time, and Since there is no search omission, there is an effect that productivity is improved and high reliability can be maintained.

Further, by learning, the size of the search area is changed and what size of the search area detects the position of the recognition target object on average in the shortest time. Thus, the position of the recognition target can be detected without omission of the search, and the productivity is improved.

[Brief description of drawings]

FIG. 1 is a block diagram showing an example of a configuration of electronic circuit mounting equipment using an image recognition method of the present invention.

FIG. 2 is a flowchart showing an example of the teaching process of the present invention.

FIG. 3 is a flowchart showing steps of the first embodiment of the present invention.

FIG. 4 is a flowchart showing steps of a second embodiment of the present invention.

FIG. 5 is a diagram showing an example of a divided search area of the present invention.

FIG. 6 is a diagram showing another example of a divided search area of the present invention.

[Explanation of symbols]

 1 Main body of image recognition device 2 Camera 3 Monitor TV 4 Object to be recognized 5 Robot 6 Controller 7 Operation panel 8 Monitor screen 9 Search area 10 Detection position of alignment point

Front page continued (72) Inventor Takao Nagumo 1006 Kadoma, Kadoma City, Osaka Prefecture Matsushita Electric Industrial Co., Ltd.

Claims (3)

[Claims] 1. A method of recognizing an image by template matching, the teaching step of dividing an image to be searched into a plurality of search areas, and teaching each search area in a predetermined search order; A recognition step of searching the image position of the recognition target object based on the order, a recording step of recording the recognition result, and a predetermined number of times of recognition, and the recognition target image is detected in descending order of frequency of recognition based on past recognition results. An image recognition method comprising: a learning step of learning so as to arrange search areas; and a search order changing step of rearranging the previously taught search order in the learned order. 2. An image recognition method using template matching, in which an image to be searched is divided into a plurality of search areas, and a teaching step of teaching to search each search area in a predetermined search order; The recognition step of searching the image position of the recognition target object based on the order, the recording step of recording the recognition result, and the search area previously taught after the predetermined recognition number of times, the past recognition target image detection position. First learning step of learning so as to match the statistical distribution, and after a predetermined number of recognitions, the search area set according to the statistical distribution has a high detection frequency of the recognition target image based on the past recognition result. The second learning step of learning so as to be arranged in order, and the previously taught search area and search order are the learned search area and search. An image recognition method comprising: a search area and a search order changing step of replacing the order with the order. 3. The image recognition method according to claim 2, wherein the search area learned in the first learning step is gradually made larger or smaller so that the average recognition time is the largest in the past recognition results. An image recognition method characterized by adding a third learning step of learning a size of a search area which becomes shorter and a search order.