JP2008151607A - Method and apparatus for positioning mark using template matching - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To carry out positioning precisely without the need for processing other than template matching for performing the positioning minutely, even in the case where the size of a recognized mark differs from that of a mark on a template image. <P>SOLUTION: The mark positioning uses the template matching, wherein a template image including the mark is moved sequentially on a recognized image, and the maximum correlation position is obtained. The mark positioning grasps a peak overall shape, defines a narrow range around the peak top in the case of a sharp peak, or a wide range containing the whole area of the flat top in the case of a flat peak, as a sampling area, and calculates a detection position from a weighted average of correlation values in the sampling area. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to a mark positioning method and apparatus using template matching for sequentially moving a template image including a mark on a recognition image to obtain a maximum correlation position, and in particular, when positioning a mark with a component mounting machine. The present invention relates to a mark positioning method and apparatus capable of accurately detecting the center position of a circular or polygonal mark.

  As a method for positioning various alignment marks arranged on a printed circuit board, a method is generally used in which a mark is picked up by a camera and subjected to some image processing to detect the mark position.

  As a mark positioning method, for example, there is a method using template matching. This is a technique for obtaining the maximum correlation position by sequentially moving the template image including the mark on the recognition image. However, since the correlation can be obtained only in the pixel unit, the positioning is performed in the sub-pixel unit smaller than the pixel. Many methods have been proposed (Patent Documents 1, 2, and 3).

  When positioning a figure such as a mark by template matching, it is necessary to create a template image having the same shape and size as the image to be recognized. If there is a difference between the figure to be recognized and the figure on the template image, it is difficult to peak the correlation position.

  For example, in the case of a circular mark, the difference in correlation distribution between the case where the recognition mark and the mark on the template are the same size and the case where there is a deviation in size are illustrated in FIG. As shown in FIG. 1A, when the size of the recognition mark coincides with the mark on the template, the peak is highly sharp at the maximum correlation position, whereas as shown in FIG. When the mark sizes do not match, the peak is low and flat. Furthermore, as shown in an enlarged view in the lower part of FIG. 1B, small unevenness exists in the flat portion due to the density unevenness of the recognized image and the influence of noise. Even if the maximum correlation position is detected in this state and positioning is performed with sub-pixel accuracy by the previous method, it only detects a small local peak, and is greatly different from the mark position to be actually obtained. ing. Also, the presence or absence of density unevenness or noise varies depending on the photographed image, which may cause the detection position to vary greatly.

  In the case of mark positioning of component mounters, the mark size varies widely depending on the board, or the same mark is recognized by multiple cameras, so even if the same mark is recognized, the mark size varies with each recognition There is. For this reason, the positioning accuracy of the mark is not stable, which is a problem.

  With respect to such a problem, in Patent Document 4, the positioning by template matching is rough positioning, the mark edge is detected in the vicinity of the coarse positioning, the quality of the detected edge is judged, and the mark size or the like is determined only from the good edge. A method for obtaining the position has been proposed.

  In Patent Document 5, pyramid matching is used to increase the resolution step by step until the same-size image is obtained by setting a matching area with the next resolution from the peak information detected at the previous resolution. It is described to do.

JP-A-4-115376 Japanese Patent No. 2943257 Japanese Patent Laid-Open No. 61-260235 Japanese Patent No. 3520758 JP 2001-148014 A (paragraph [0005] FIG. 8)

  However, in the method of Patent Document 4, it is necessary to switch the algorithms for edge quality determination, mark size acquisition, and mark position detection depending on the mark shape. Further, in addition to template matching, there is a problem that processing time is required because processing for detailed positioning is required.

  Moreover, only the method of patent document 5 cannot solve the problem similar to patent documents 1-3.

  The present invention has been made to solve the above-described conventional problems, and requires processing for detailed positioning in addition to template matching even if the size of the mark to be recognized differs from the size of the mark on the template image. Therefore, it is an object to enable positioning with high accuracy.

  The present invention is a mark positioning method using template matching that sequentially moves a template image including a mark on a recognition image to obtain a maximum correlation position. The above-mentioned problem has been solved by calculating the detection position from the weighted average of the correlation values in the sampling region, with a narrow range around the periphery, and a wide range including all the flat part at the top in the case of a flat peak as the sampling region. Is.

  Also, in the template matching, pyramid matching is performed in which the resolution is increased step by step until the same-magnification image is obtained by setting a matching area at the next resolution from the peak information detected at the previous resolution. be able to.

  In the mark positioning apparatus using template matching that sequentially moves the template image including the mark on the recognition image and obtains the maximum correlation position, the present invention has a means for grasping the overall shape of the peak and the sharp peak. Means for setting a narrow range around the top of the peak, a wide range including all of the top of the peak if it is a flat peak, and means for calculating the detection position from a weighted average of correlation values in the sampling region; And a mark positioning device using template matching characterized by comprising:

  According to the present invention, even if the size of the mark to be recognized and the size of the mark on the template image are different, it is possible to perform positioning accurately without requiring processing for detailed positioning in addition to template matching. .

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

  FIG. 2 shows a component mounter as an example to which the present invention is applied. In the figure, the suction nozzle 1a of the suction head 1 is driven by an XY drive mechanism (not shown), moves to the position of the feeder 2, sucks the component 3 supplied from the feeder 2, and then recognizes the camera 6. Move up. Therefore, the component 3 is picked up from below, the image is processed, the suction position deviation of the suction nozzle 1a of the component 3 is corrected, and the component is mounted at a predetermined position on the substrate 10 transported through the transport path 11. Is done. Since accurate mounting of the component 3 is based on the premise that the substrate 10 is at the reference position, substrate marks 10 a, 10 b, 10 c are formed at predetermined positions on the substrate 10 and attached to the suction head 1. (Image reading means) 12 picks up an image from above and recognizes the image, finds the deviation of the substrate 10 from the reference position, corrects this, and mounts the components.

  FIG. 3 shows a configuration of an image processing apparatus 20 that processes an image of a substrate mark picked up by the CCD camera 12. The analog image signal from the CCD camera 12 is converted into digital image data of a predetermined format by the A / D converter 20a of the image processing apparatus 20. The image data output from the A / D converter 20a is temporarily stored in the image memory 20b and returned to analog data via the D / A converter 20c as necessary (or after a predetermined image processing), It can be displayed on the monitor 21.

  Each unit of the image processing apparatus 20 is controlled by the CPU 20d. The CPU 20d stores a ROM 20e storing various programs, a RAM 20f used as a work area of the CPU 20d, and the shape and size of a mark to be recognized. A mark data area 20g and a production template memory 20h are connected. Further, the CPU 20d can control other devices, for example, a transport system for transporting the substrate 10 of the component mounting machine, the suction head 1 for attracting components, and the like according to the result of image processing via the controller 22 of each unit. It has become.

  In such a configuration, a procedure for acquiring (teaching) a model template suitable for a substrate mark and recognizing the mark will be described with reference to the flowcharts of FIGS.

  In teaching, as shown in FIG. 4, first, the suction head 1 is moved to, for example, the area of the substrate mark 10a, and the substrate mark 10a is read by the CCD camera 12 attached to the suction head 1, and the A / D converter It is converted into digital data at 20a and stored in the image memory 20b (step S1). Subsequently, an area to be subjected to the matching process on the image data in the image memory 20b, that is, a matching window W is set (step S2). The setting of the matching window W is to roughly detect the position of the substrate mark 10a from the mark region on the substrate by searching density information, etc., and specify the matching processing region. It does not require accuracy.

  Subsequently, features such as a density change curve, its inflection point, or the position of these pieces of information in the matching window W are extracted from the image data in the matching window W (step S3). Here, at least the minimum necessary feature portion is extracted for recognizing the substrate mark 10a using a model template to be described later. For example, the projection data, size, and density information maximum / minimum of the substrate mark 10a are extracted. Value features are extracted.

  Next, in step S4, a model template is created. For example, in the case of corresponding to the shapes of 31 to 39 in FIG. 6, first, the gradation density of 31 is first determined based on the characteristics of the image extracted in step S3, that is, the maximum density, minimum density, and mark size of the image. A two-dimensional model template having data is generated.

  The correlation between the model template thus generated and the image data of the matching window W is calculated (step S5), and the matching result, for example, the matching coefficient (correlation coefficient) is stored in the RAM 20f together with the generated model template and shape name. (Step S6).

  Next, model templates are created for the other model shapes 32 to 39, and the above process is repeated. When model templates are generated for all shapes and the above process is completed (YES in step S7), it is searched whether there is a model template suitable for the board mark 10a (step S8). This is done by displaying the matching result on the monitor 21 and checking whether there is a model template whose matching coefficient is larger than the set value. If there is a model template larger than the set value (YES in step S8), the one with the largest matching coefficient is selected as the model template for the substrate mark 10a (step S9).

  On the other hand, if there is no matching coefficient larger than the set value in step S8 and there is no suitable template, a template list is displayed on the monitor 21, and the model for the substrate mark 10a is checked by the operator from among them. A template is selected (step S10).

  Next, the above processing is performed on the substrate marks 10b and 10c (step S12). As a result, teaching of all substrate marks on the substrate 10 has been performed.

  The mark shape and the mark size obtained by the teaching operation are given an ID for each recognition mark together with the mark search range, the expected matching coefficient, etc., and are stored in the mark data area 20g (step S13).

  At the time of board production, as shown in FIG. 5, first, a board type (ID) to be produced (mounted) is input (step S20). Subsequently, a model template corresponding to each board mark of the board type is generated from the mark data stored in the mark data area 20g (step S21). The generated model template is stored in the production template memory 20h. Here, if there is a margin in the RAM 20f, a model template can be generated at the stage where the mark data is stored in step S13 of FIG. 4 and stored in advance in the RAM 20f.

  When the substrate 10 enters the conveyance path 11 and is carried along (step S22), the suction head 1 moves to each mark area where the substrate mark is formed, and an image of each substrate mark is obtained by the CCD camera 12. Each is read (step 23). Subsequently, template matching is performed using the model template corresponding to the substrate mark generated in step S21, and each substrate mark is recognized (step S24). The present invention is for accurately performing template matching in this step, and details will be described later.

  According to the result of this mark recognition, a correction value of the substrate is calculated (step S25), the transport path 11 is controlled via the controller 22, the position of the substrate 10 is corrected, and the substrate 10 is positioned (step S26). Subsequently, component mounting is performed (step S27).

  The present invention provides template matching that is effective for both high and sharp peaks as shown in FIG. 1A and low and flat peaks as shown in FIG. . A schematic diagram (two-dimensional representation) of the positioning method is shown in FIG.

  The point is not to perform sub-pixel calculation using local calculation around the maximum correlation position, but to grasp the whole peak shape, if it is a sharp peak, a narrow range at the top of the peak, if it is a flat peak, The detection position is calculated from the weighted average of the correlation values in the sampling area, with a wide range including the flat part of the sampling area as a sampling area.

  In template matching, pyramid matching is generally used to reduce processing time. This is because, as described in Patent Document 5, before performing template matching with the same-size image, the resolution is reduced in several steps, and the candidate position is first narrowed down with the image with the coarsest resolution. In this method, matching calculation is performed on the next resolution image, and positioning is finally performed on the same-size image.

  In the present embodiment, the present invention will be described in combination with this pyramid calculation. FIG. 8 shows the procedure of this embodiment.

  First, a pyramid image with reduced resolution is generated for both the recognition image and the template image (step S30). The extent to which the coarsest resolution is reduced is determined according to the size of the image and the size of the recognition mark.

  Next, a matching operation is performed on the recognition image with the lowest resolution and the template image (step S31). The correlation value distribution obtained here is, for example, as shown in FIG. 9A. By providing a certain appropriate threshold, a peak exceeding this threshold is extracted as shown in black in FIG. 9B. (Step S32). The threshold value used at this time may be obtained as a percentage of the maximum correlation value. When the correlation value of the mark expected by the user is known, the threshold value may be determined based on this value. The circumscribed rectangle of the peak detected here is obtained as illustrated in FIG.

  Thereafter, the resolution is increased stepwise until an equal-magnification image is obtained (step S33), and matching calculation is performed only in the vicinity of each peak detected in the previous layer. That is, the next matching area is set from the peak information detected at the previous resolution (step S34), and the peak is set in the rectangle surrounding the peak (the circumscribed rectangle obtained when calculating the peak or the previous matching area). The center position and kurtosis of the peak (peak spread amount) are calculated, and using this, a matching area is set, and template matching is performed within the matching area thus obtained (step S35).

  Specifically, as illustrated in FIG. 11 as a two-dimensional example, the circumscribed rectangle of the lump obtained using the maximum correlation value thr on the four sides (Sp and Ep) of the rectangle A surrounding the peak as a threshold is calculated as the center calculation. Set as rectangle B.

  Next, in the center calculation rectangle B, as shown in the following equation, a weighted average of the correlation distribution that weights the correlation value V (x) with respect to the coordinate x is performed, and the peak center position Cp is calculated.

  Thereby, the peak center can be calculated without being affected by the unevenness of the peak flat portion.

  Next, the correlation distribution around the peak is radiated from the peak center position Cp, and a position that greatly fluctuates with respect to the correlation value at the peak center (for example, 5% fluctuation of the peak correlation position) is calculated. At this time, the correlation distribution may be smoothed.

  Next, the rectangle D surrounding the previously obtained variation position is set as the next matching region.

  When matching is completed for all peaks detected in the coarsest layer (YES in step S36), the resolution is increased (step S33), and steps S34 and S35 are repeated.

  When matching with the same size image is completed (YES in step S37), the highest peak among the detected peaks is selected (step S38), the peak center position is calculated (step S39), and this is finally detected. Position.

  In this embodiment, since pyramid matching is used together, the peak center position can be accurately obtained by simple calculation. It should be noted that template matching according to the present invention can also be performed directly on an equal-size image without performing pyramid matching.

  In the said embodiment, although this invention was applied to the component mounting machine, the application object of this invention is not limited to this.

Illustration for explaining conventional problems The perspective view which shows the principal part structure of the component mounting machine which is an example of the application object of this invention Block diagram showing the configuration of the image processing apparatus Flow chart showing the procedure for obtaining (teaching) model templates that are also suitable for board marks Flow chart showing the procedure for mark recognition Similarly, a plan view showing an example of a substrate mark The figure which shows the concept of the template matching by this invention The flowchart which shows the process sequence in embodiment of this invention The figure which similarly shows the example of distribution of correlation value A plan view showing the circumscribed rectangle The figure which shows the detail of the process sequence by this invention

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 ... Adsorption head 2 ... Feeder 3 ... Parts 10 ... Substrate 10a, 10b, 10c ... Substrate mark 11 ... Conveyance path 12 ... CCD camera 20 ... Image processing device 22 ... Controller

Claims (3)

In the mark positioning method using the template matching for sequentially moving the template image including the mark on the recognition image and obtaining the maximum correlation position,
Understand the overall shape of the peak,
If it is a sharp peak, a narrow range around the top of the peak, and if it is a flat peak, a wide range including all the flat part of the top is used as a sampling region.
A mark positioning method using template matching, wherein a detection position is calculated from a weighted average of correlation values in the sampling region.   In the template matching, by setting a matching area at the next resolution from the peak information detected at the previous resolution, pyramid matching is performed in which the resolution is increased step by step until an equal-magnification image is obtained. The mark positioning method using template matching according to claim 1. In the mark positioning device using template matching for sequentially moving the template image including the mark on the recognition image and obtaining the maximum correlation position,
Means to understand the overall shape of the peak;
If it is a sharp peak, a narrow range around the peak top, and if it is a flat peak, a means to set the sampling range as a wide range including all of the flat top,
Means for calculating a detection position from a weighted average of correlation values in the sampling region;
A mark positioning device using template matching characterized by comprising:

Images