JP3520758B2 - Image recognition method - Google Patents

Description

Description: BACKGROUND OF THE INVENTION [0001] 1. Field of the Invention [0002] The present invention relates to an image recognition method for recognizing a recognition mark such as an electronic component on an image. 2. Description of the Related Art As a positioning method in an electronic component mounting apparatus or the like, a method using image recognition is widely used. This method takes an image of a range including a recognition mark provided on an electronic component or a board, specifies the recognition mark from the captured image, and detects the position of the specified recognition mark to thereby detect the electronic component or the board. The position is determined.
When this detection is performed, a registered pattern registered in advance in the image recognition apparatus as a pattern image of the recognition mark is matched with the captured image. In the conventional image recognition method, at the time of registration of a registration pattern, a teaching operation of selecting an electronic component or a board to be inspected having a good recognition mark from the electronic component and a board and taking the recognition mark with a camera is required. [0003] However, since the above-mentioned teaching operation requires skill, a registration error is likely to occur at the time of capturing a template image, and the registration error is directly detected at the detection position coordinates. Error. That is, there is a problem that the detection accuracy of the recognition mark is restricted by the skill of the operator who performs the teaching operation at the time of registering the pattern. Further, in the conventional image recognition method, the entire recognition mark is recognized as a lump, and matching is determined based on the luminance of each part of the recognition mark. If the size of the mark varies, there is a problem that the recognition accuracy is not stable and a recognition error is likely to occur. As described above, the conventional image recognition method has a problem that the skill is required at the time of pattern registration, the operability is poor, and the recognition accuracy is not stable. Accordingly, an object of the present invention is to provide an image recognition method which has excellent operability and stable recognition accuracy. [0005] An image recognition method according to the present invention stores image data including a recognition mark captured by a camera in an image storage unit, and registers the recognition mark stored in the registration pattern storage unit. An image recognition method for identifying the recognition mark by referring to a pattern and the image data and detecting a position coordinate of the recognition mark, wherein a step of inputting feature data of the registered pattern, and a step of inputting the captured recognition mark a step of storing the image data in the image storage unit which includes a step of storing the registration pattern storage unit to generate a registered pattern based on the feature data, the pattern matching
A step of estimating the recognition mark looking for images similar to the registration pattern from the image in the image storage unit Ri, the step of measuring the edge of the estimated recognition mark images,
A step of determining the quality of the measured edge data; and a step of obtaining the position coordinates of the recognition mark based on the edge data determined to be good. According to the present invention, a registration pattern is generated based on characteristic data of a registration pattern for image recognition, an image similar to the registration pattern is obtained from a captured image, and edges of the image are measured. By determining whether the edge is good or bad, it is possible to obtain stable recognition accuracy with good operability even if the recognition mark has dirt or partial chipping. Next, an embodiment of the present invention will be described with reference to the drawings. FIG. 1 is a block diagram showing the configuration of an image recognition device according to an embodiment of the present invention, FIG. 2 is a flowchart showing the image recognition method, FIGS. 3, 4, and 5 are image diagrams of the recognition mark, FIG. FIG. 7 is a perspective view showing a three-dimensional model of a luminance distribution of an edge portion of the recognition mark, FIG. 7 is a block diagram showing a method of determining whether or not the edge of the recognition mark is good, and FIG. 8 is a display screen of the image recognition device. First, the configuration of the image recognition device will be described with reference to FIG. In FIG. 1, a camera 1 captures an image of a range including a recognition mark M provided on a substrate 10 below. The AD converter 2 converts analog image data captured by the camera 1 into digital image data. The image storage unit 3 stores the converted image data. Registration pattern storage unit 4
Stores a pattern of a figure registered as a detection target. The operation data input unit 5 includes a keyboard, a touch panel,
The CPU 7 includes a pointing device and the like.
Input the feature data of the registration pattern automatically generated in. The CPU 7 generates a registration pattern based on the input feature data according to the program stored in the program storage unit 6 and stores the registration pattern in the registration pattern storage unit 4. The display unit 8 is a CRT and displays an image of the recognition mark M. Next, an image recognition method will be described with reference to the drawings according to the flow of FIG. First, a pixel rate is input as a preparation operation. This operation involves inputting a unique value (representing the correspondence between the pixel and the actual size) that differs for each device at the time of starting the device, whereby the actual size of the imaging target can be obtained on the screen. . Next, when the type of the board 10 to be image-recognized is specified, the feature data of the registration pattern of the recognition mark M used for the board 10 is input from the operation data input unit 5. FIG. 8 shows a display screen at the time of inputting feature data. A feature data window 30 is displayed on the screen of the display unit 8, and shape selection buttons 31, 3
2, 33 are arranged. When a shape to be registered is selected from the shape buttons with the pointer 34, a sub-window 35 is displayed. FIG. 8 shows a case where a circle is selected as the shape. Then, text boxes 36 and 37 in the window 35 are input with information on dimensions and colors (white and black), and a registration button 38 is selected with the pointer 34 to complete the registration work. The feature data input in the feature data input window is stored in the registered pattern storage unit 4. In this embodiment, an example is shown in which a white circle having a diameter of 0.1 mm is used as the recognition mark M.
In this way, when registering a pattern image, it is only necessary to simply input the characteristic data, and there is no need for a conventionally required operation such as a teaching operation requiring skill, and the operability at the time of pattern registration is significantly improved. be able to. Next, a flow of recognizing a recognition mark M on the substrate 10 to be positioned on an image and detecting a position will be described. First, a range including the recognition mark M of the substrate 10 is imaged by the camera 1, and the image data obtained is taken into the image storage unit 3 (ST1). In parallel with this, a registration pattern is generated based on the feature data of the registration pattern input in the CPU 7 (ST2). Next, an image similar to the registered pattern is searched for from images captured and taken into the image storage unit 3 using pattern matching processing (ST3). That is, the image of the recognition mark M is estimated from the images in the imaging range. At this time, as shown in FIG. 3, the size of the similar image M ′ selected from the actually captured image does not always match the size of the registered pattern R generated from the feature data. , The diameter may be different and the center position may be shifted. This is because the recognition mark M formed on the substrate 10 by a method such as printing does not always completely match the feature data due to an error in forming the mark. Next, if a similar image is obtained, that is, if the recognition mark M is estimated, the edge of the similar image is measured (ST4). At this time, if the recognition mark M is correctly formed in a circular shape as shown in FIG. 4A, the similar image M ′ is a normal circular shape, and the edge point Pi is obtained on the circumference of a substantially correct shape. Can be On the other hand, when the recognition mark M is partially missing, the similar image M ′ also has a missing shape as shown in FIG.
i is obtained not only on the original circumference but also at a position deviating from the circumference (see Pi ′ surrounded by a broken line A). Next, in order to eliminate the influence caused by the defective formation of the recognition mark M, the quality of the edge data is determined (ST5). Here, since the feature data of the pattern image has already been input and the shape of the recognition mark M has been specified on the assumption that it is a circle, the edge points deviating from the circumference as shown in FIG. P
i ′ is rejected as bad data. Then, only the center point of the circumference and the size (radius r) of the similar image (recognition mark) are obtained from only the edge points determined as good data (ST).
6). At this time, since the center point C is obtained based only on the edge data of the similar image, even if there is unevenness inside the similar image as shown in FIG. 5A, the similarity is obtained as shown in FIG. Even if the image is partially missing, the accuracy of the required center position is not affected. Next, the reliability of the similar image is evaluated based on the center position data (center point C) and size data (radius r) obtained above (ST7). Hereinafter, this evaluation method will be described. First, the three-dimensional model of the luminance distribution shown in FIG. 6 is obtained by extracting an edge portion of an ideal recognition mark M as a circumferential element, dividing it into n parts, and modeling one part thereof. The height of the annular portion 11 shown in FIG. 6 indicates the luminance of the edge, and the width b of the annular portion 11 corresponds to the width of several pixels. As described above, since the recognition mark M is white, the luminance h1 on the imaging screen is higher than the image on the upper surface of the substrate 10 and the height h1 indicating the luminance of the annular portion 11 of the edge is:
It appears higher than the height h2 indicating the brightness of the annular flange 12 (the brightness of the upper surface of the substrate). That is, the ideal recognition mark M
The brightness distribution model of the edge portion of the image shown in FIG. That is, the edge is a perfect circle,
The inside of the circle inside the edge is in a state of being uniformly white-painted. The evaluation of the reliability of the similar image is performed by comparing the ideal edge with the actual edge. In FIG. 7, reference numeral 20 denotes the ideal edge model, and reference numeral 21 denotes a luminance distribution of a circumferential element cut out from an actual similar image M ′ based on the center position data and the size data obtained in ST6. Model. By comparing the three-dimensional shapes of the model 20 and the model 21,
The dimensional coincidence is compared by the coincidence calculator 22. Then, based on the degree of coincidence determined here, the edge determination section 23 determines whether or not the portion is an edge. Then, by integrating the data of the determination result of each of the divided edges with respect to the entire circumference in the integrator 24, the ratio of the portion determined to be the edge to the entire circumference, that is, the good edge ratio can be obtained. If the good edge ratio is equal to or greater than a preset threshold,
It is determined that there is no error in the reliability of the similar image, that is, the estimation of the recognition mark and the state of the recognition mark is good, and the result of the image recognition is passed (ST).
8). As described above, in the image recognition method in which the recognition mark provided on the substrate is image-recognized and the position is detected, only the edge of the pattern of the recognition mark is recognized on the image, the dimension is measured, and Determining the center position enables high-accuracy position detection even when there is printing unevenness or chipping of the recognition mark, which had affected recognition accuracy when the image was recognized as a lump as in the past. Can be. According to the present invention, a registration pattern is generated based on feature data of a registration pattern for image recognition, an image similar to the registration pattern is obtained from a captured image, and an edge of the image is obtained. Is measured and the quality of the edge is determined, so that the teaching operation, which required skill in the past, can be omitted and pattern registration can be performed only by data input, greatly improving the operability at the time of pattern registration. Can be improved. Also, since only the edges of the pattern of the recognition mark are recognized on the image, position detection can be performed with high accuracy even when there is unevenness or partial chipping inside the recognition mark.

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a block diagram illustrating a configuration of an image recognition device according to an embodiment of the present invention. FIG. 2 is a flowchart illustrating an image recognition method according to an embodiment of the present invention. FIG. 4 is an image diagram of a recognition mark according to an embodiment of the present invention. FIG. 5 is an image diagram of a recognition mark according to an embodiment of the present invention. FIG. 7 is a perspective view showing a three-dimensional model of the luminance distribution of the edge portion of the recognition mark according to one embodiment of the present invention. FIG. 8 is a diagram showing a display screen in the image recognition device according to one embodiment of the present invention. [Description of References] 1 Camera 3 Image storage unit 4 Registration pattern storage unit 5 Operation data input unit 6 Program storage unit 10 Substrate M recognition mark R registration pattern

Claims (1)

(57) Claims 1. An image data including a recognition mark captured by a camera is stored in an image storage unit, and a registration pattern of the recognition mark stored in the registration pattern storage unit and the image data are stored in the storage unit. An image recognition method for identifying the recognition mark with reference to detecting the position coordinates of the recognition mark, comprising: inputting feature data of the registration pattern; and storing image data including the captured recognition mark in an image. a step of storing the parts, a step of storing the registration pattern storage unit to generate a registered pattern based on the feature data, pattern matching
A step of estimating the recognition mark looking for images similar to the registration pattern from the image in the image storage unit by ring, the step of measuring the edge of the image of the estimated recognition marks, the edge data measured An image recognizing method, comprising: a step of determining pass / fail; and a step of obtaining position coordinates of the recognition mark based on edge data determined to be pass / fail.