JP3442163B2 - Positioning method and apparatus - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a positioning method and apparatus suitable for positioning printed boards. 2. Description of the Related Art In recent years, in order to realize miniaturization and high functionality of information equipment, multilayering and increasing the density of printed circuit boards have been accelerated. As a result, the substrate pattern is miniaturized, and there is a strong demand for the development of an apparatus that automatically and accurately performs precise positioning in component mounting.

[0002]

2. Description of the Related Art Conventionally, as a position recognition means for position alignment, for example, as shown in Japanese Patent Laid-Open No. 4-307682, a reference image for position alignment is used as a reference image, and light and shade pattern matching is performed. A method of performing registration by calculating a normalized correlation value or the like with the search image and recognizing the position over the entire target image has been widely used.

The above-mentioned conventional technique has a problem that the processing takes too long when the reference image and the target image are large. Therefore, it is necessary to take measures such as roughening the pitch for obtaining the normalized correlation value (reducing the calculation points).
It had a negative effect on accuracy. Therefore, there has been proposed a method of performing a rough alignment and a fine alignment, and performing a high-speed and precise alignment even when the reference image and the target image are large (see, for example, Japanese Patent Laid-Open No. 3-225481). ).

32 and 33 are flowcharts showing the processing procedure of the above-mentioned conventional technique. In the figure, a target image is input in step S1, and the input image is binarized in step S2. Then, in step S3, a binary projection pattern for each of the two XY directions is created. On the other hand, in step S4, design data (which serves as a reference image) is input, in step S5 the reference image data is binarized, and in step S6, XY
A projection pattern is created for each of the two directions.

In step S7, the rough positional deviation amount is calculated from the position where the residual difference between the binary projection patterns of the target image and the reference image created in steps S3 and S6 becomes the minimum, and in step S8, the rough registration is performed. I do. In step S9, the binary target image that has been roughly aligned and the binary reference image are used to perform two-dimensional pattern matching by the residual test method shown in FIG.

Then, in step S10, based on the result calculated in step S9, the XY stage is moved to perform precise alignment of the object.

[0007]

The above-mentioned prior art has the following problems. Since the pattern matching is performed by binarizing the target image and the reference image, it is not possible to perform highly accurate alignment. The rough alignment of the target image and the reference image is performed by using two projection patterns in the XY directions, and the fine alignment is performed by narrowing down the candidate points to one in the rough alignment. If the alignment is wrong,
It is necessary to start the process over from the beginning, resulting in poor processing efficiency.

The present invention has been made in consideration of the above-mentioned problems of the prior art, and a first object of the present invention is to perform a high-speed alignment without lowering the positioning accuracy. A method and apparatus are provided. It is a second object of the present invention to provide a positioning method and device that make it easy to grasp the rough position and perform accurate rough positioning.

A third object of the present invention is to provide a positioning method and apparatus which can reduce the possibility of erroneous recognition with respect to noise, changes in illumination, etc. and can perform highly accurate positioning.

[0010]

FIG. 1 shows the principle of the present invention. In the figure, 1 is an alignment target object,
A mark, such as a cross shape, which serves as a reference for alignment, is printed on the object. 2 is an image input device for taking in the target image of the object, 3 is an image memory for storing the input image data, and 4 is a candidate for extracting a candidate point for positioning from the image data stored in the image memory 3. Point extracting means, 5 is rough positioning means for performing rough positioning from the extracted candidate points, 6 is fine positioning means for performing finer positioning of the roughly positioned candidate points, and 7 is for controlling each of the above means. It is an overall control means.

In order to solve the above problems, the invention according to claim 1 of the present invention captures a target image on a member on which a mark to be aligned is marked, and a one-dimensional projection distribution of the captured image in the XY directions. From this, one or a plurality of candidate points for alignment are extracted, and the extracted one or a plurality of candidate points are subjected to light and shade pattern matching with a reference image to obtain a rough position for alignment. In the area near the rough position, the light and shade pattern matching with the reference image is performed to perform the fine alignment.

According to a second aspect of the present invention, in an apparatus for aligning an object at a predetermined position, an image input means for taking in a target image on a member marked with a mark to be aligned, and taking in the image. Image memory that stores image data and the above image memory are accessed and imported
Candidate point extracting means for extracting one or a plurality of alignment candidate points from the one-dimensional projection distribution in the XY directions of the image, and rough positioning means for roughly aligning the extracted one or a plurality of candidate points. Provided with a precision positioning means for performing a finer alignment by performing a light and shade pattern matching with the reference image only in the rough position vicinity area obtained by the rough positioning means, and an overall control means for controlling each of the above means Is.

According to the invention of claim 3 of the present invention, in the invention of claim 2, a pattern which is optimum for recognizing a target image having a shape and density corresponding to the pattern to be recognized is used as the reference image.

[0014] The invention of claim 4 of the present invention is also Claim 2
According to the third aspect of the invention, the candidate point extracting means is configured to change the condition for rough position extraction according to the brightness of the target image due to the change in illumination. According to a fifth aspect of the present invention, in the second, third or fourth aspect of the invention, the coarse positioning means matches the grayscale pattern with the reference image for each of the extracted one or a plurality of candidate points. Is performed, and the coordinates having the highest correlation value are set as the coarse position.

The invention of claim 6 of the present invention is the invention of claim 2
According to the third or fourth aspect of the present invention, the fine positioning means is configured to perform a spiral search from the center of the rough position vicinity region toward the outside when performing pattern matching with the reference image. According to a seventh aspect of the present invention, in the second, third, fourth, fifth or sixth aspect of the invention, when pattern matching with a reference image is performed, an initially operable one is calculated in advance, It is configured to be processed as a constant.

The invention of claim 8 of the present invention is the invention of claim 2
In the third, fourth, fifth, sixth or seventh aspect of the invention, when performing pattern matching with the reference image, the weight around the contour of the target image data is reduced to obtain the correlation value. The invention of claim 9 of the present invention is
2, 3, 4, 5, 6, 7 or in the invention of claim 8 , when the sizes of the reference image and the target image are different from each other, the size of the reference image is automatically adjusted to determine the sizes of the reference image and the target image. It is configured to match and perform pattern matching.

The invention of claim 10 of the present invention is the invention of claim 2
In the invention of claim 3, 4, 5, 6, 7, 8 or claim 9 , a memory for storing the reference image, a memory for storing the target image, and a selector are provided, and the image captured from the image input device is selected by the selector. It can be stored in each memory. According to the invention of claim 11 of the present invention, in the invention of claim 10 , when the reference image data is read,
This is configured so that the memory storing the reference image is directly accessed.

The invention of claim 12 of the present invention is the invention of claim 2,
In the invention of 3, 4, 5, 6, 7, 8, 9, 10 or claim 11 , a memory for storing each image data of R, G, B is provided, and the target image is input as a color image, Store each R, G, B data in the above memory,
The image data stored in the R, G, and B memories are separately processed by the same algorithm, and the recognition result is obtained from the average of the above three processing results.

The invention of claim 13 of the present invention is the invention of claim 2
In the invention of 3, 4, 5, 6, 7, 8, 9, 10 or claim 11 , a memory for storing each image data of R, G, B is provided, and the target image is input as a color image, Store each R, G, B data in the above memory,
The image data stored in the R, G, and B memories are separately processed by the same algorithm, and when the difference between the above three processing results is within a certain range, the recognition result is obtained as a correct result. Is configured to output.

The invention of claim 14 of the present invention is the invention of claim 2
3, 4, 5, 6, 7, 8, 9, 10 or the invention of claim 11 is provided with a plurality of memories for storing target image data, stores the target image in the plurality of memories, and stores in each memory. The processed image data is separately processed by different algorithms, and the recognition result is obtained from the average of the three processing results.

The invention of claim 15 of the present invention is the invention of claim 2
3, 4, 5, 6, 7, 8, 9, 10 or the invention of claim 11 is provided with a plurality of memories for storing target image data, stores the target image in the plurality of memories, and stores in each memory. The processed image data is separately processed by different algorithms, and when the difference between the three processing results is within a certain range, the recognition result is output as if the correct result was obtained.

The invention of claim 16 of the present invention is the invention of claim 2
In the invention of 3, 4, 5, 6, 7, 8, 9, 10 or claim 11 , a plurality of systems for performing recognition processing are provided,
The same target image is input to the above-mentioned plurality of systems, each system is separately processed by different algorithms, and the recognition result is obtained from the average of the above-mentioned three processing results.

The invention of claim 17 of the present invention is the invention of claim 2
In the invention of 3, 4, 5, 6, 7, 8, 9, 10 or claim 11 , a plurality of systems for performing recognition processing are provided,
The same target image is input to the above-mentioned plurality of systems, processed differently by each system by different algorithms, and a recognition result is obtained when a correct result is obtained when the difference between the above-mentioned three processing results is within a certain range. Is configured to output.

The invention of claim 18 of the present invention is the invention of claim 2
In the invention of 3, 4, 5, 6, 7, 8, 9, 10 or claim 11, a plurality of systems for performing recognition processing are provided,
The same target image is input to the above-mentioned plurality of systems, processed differently by each system by different algorithms, and the fastest processed result of the above three processed results is output as a recognition result. It was done.

The invention of claim 19 of the present invention is the invention of claim 2,
In the invention of 3, 4, 5, 6, 7, 8, 9, 10 or claim 11 , a plurality of systems for performing recognition processing are provided,
One target image is divided into a plurality of parts, and the plurality of divided target images are processed in parallel by the system to obtain a recognition result. The invention of claim 20 of the present invention is the invention of claims 2, 3, 4, 5, 6, 7, 8, 9, 10,
11, 12, 13, 14, 15, 16, 17, 18 also
In the invention of claim 19 , there is provided trigger means for generating a trigger signal each time a predetermined amount of target image data is input to the image memory, and when the trigger means generates an output,
It is configured to start the recognition process of the target image data.

[0026]

In FIG. 1, the target image of the target object 1 captured by the image input device 2 is stored in the image memory 3. The candidate point extracting means 4 extracts one or a plurality of candidate points from the one-dimensional projection distribution of the target image stored in the image memory 3 in the XY directions. The rough positioning means 5 performs a gray pattern matching between the extracted one or a plurality of candidate points and the reference image to perform rough positioning.

The fine positioning means 6 performs the fine positioning by performing the light and shade pattern matching with the reference image in the area near the rough position. According to the first and second aspects of the present invention, as described above, the target image on the member on which the mark to be aligned is marked is captured, and the one-dimensional projection distribution of the captured image in the XY directions. From this, one or a plurality of candidate points for alignment are extracted, and the extracted one or a plurality of candidate points are subjected to light and shade pattern matching with a reference image to obtain a rough position for alignment. Since the rough pattern vicinity region is subjected to the light and shade pattern matching with the reference image to perform the fine alignment, high-speed processing can be performed without lowering the positioning accuracy. In addition, rough position candidates can be grasped by simple processing.
It can be grasped, and accurate rough position extraction can be performed.
Wear.

According to the third aspect of the present invention, since the optimum pattern for recognizing the target image having the shape and density corresponding to the pattern to be recognized is used as the reference image, the processing speed is higher than that of the rectangular pattern. Can be converted. In addition, since it is possible to cut a portion where noise is easily added, it is possible to improve accuracy.

In the invention of claim 4 of the present invention, the candidate point extracting means is configured to change the condition of the rough position extraction according to the brightness of the target image due to the change of the illumination. Accurate recognition can be performed even when the illumination changes. In the fifth aspect of the present invention, the rough positioning means performs the light and shade pattern matching with the reference image for each of the extracted one or a plurality of candidate points, and the coordinate with the highest correlation value is used as the rough position. Therefore, the possibility of erroneous recognition at the time of rough positioning is reduced, and accurate rough positioning is possible.

In the sixth aspect of the present invention, the precision positioning means is configured to perform a spiral search from the center of the rough position vicinity region to the outside when performing pattern matching with the reference image. Compared with the raster scan method, the time required to reach the point to be extracted can be shortened and the processing speed can be increased. Claims of the invention
According to the seventh aspect of the invention, when pattern matching with the reference image is performed, an initially operable one is calculated in advance,
Since the processing is performed as a constant, the time for calculating the correlation value can be shortened and the processing speed can be increased.

In the eighth aspect of the present invention, when performing pattern matching with the reference image, the weight around the contour of the target image data is reduced to obtain the correlation value. Is possible. In the invention of claim 9 of the present invention, when the reference image and the target image have different sizes, the size of the reference image is automatically adjusted,
Since the pattern matching is performed by matching the sizes of the reference image and the target image, it is possible to handle differences in magnification that are difficult for pattern matching processing to easily recognize objects with different magnifications. can do.

According to the tenth aspect of the present invention, a memory for storing the reference image, a memory for storing the target image, and a selector are provided, and the image captured from the image input device is stored in each memory by the selector. Since it is possible, it is not necessary to drop the reference image in the file, time,
You can save time. According to the eleventh aspect of the present invention, since the memory for storing the reference image data is directly accessed, the processing time can be shortened.

In the twelfth aspect of the present invention, the target image is input as a color image, stored in each memory for each R, G, B data, and stored in each memory for R, G, B. Since the image data is separately processed by the same algorithm and the recognition result is obtained from the average of the three processing results, accurate recognition can be performed.

In the thirteenth aspect of the present invention, the target image is input as a color image, stored in each memory for each R, G, B data, and stored in each memory for R, G, B. Since the image data is separately processed by the same algorithm and the recognition result is output as a correct result is obtained when the difference between the above-mentioned three processing results is within a certain range, misrecognition may occur. Accurate recognition can be performed with reduced possibility.

In the fourteenth aspect of the present invention, the target image is stored in a plurality of memories, the image data stored in each memory is processed separately by different algorithms, and the average of the three processing results is averaged. Since it is configured to obtain more recognition results, it is possible to cover the shortcomings of the respective algorithms and perform accurate position recognition. Claim 1 of the present invention
According to the fifth aspect of the invention, the target image is stored in a plurality of memories, the image data stored in each memory is processed separately by different algorithms, and the difference is correct when the difference between the three processing results is within a certain range. Since the recognition result is output as a result is obtained, it is possible to cover the shortcomings of the respective algorithms to enable accurate position recognition and reduce the possibility of erroneous recognition.

In the sixteenth aspect of the present invention, the same target image is input to a plurality of systems, each system processes it separately by a different algorithm, and the recognition result is obtained from the average of the three processing results. Since it is configured to obtain, more accurate position recognition is possible. According to a seventeenth aspect of the present invention, when the same target image is input to a plurality of systems and processed by different algorithms separately by each system, and when the difference between the three processing results is within a certain range, Since the recognition result is output assuming that the correct result is obtained, accurate position recognition is possible and the possibility of erroneous recognition can be reduced.

In the eighteenth aspect of the present invention, the same target image is input to a plurality of systems, and each system separately processes the different algorithms to obtain the fastest one of the three processing results. Since the processed result is output as the recognition result, the recognition process can be performed in a short time. In the nineteenth aspect of the present invention, one target image is divided into a plurality of parts, and the plurality of divided target images are processed in parallel by the system to obtain a recognition result. The processing can be performed in 1 time, and the processing can be speeded up.

According to the twentieth aspect of the present invention, trigger means for generating a trigger signal each time a predetermined amount of target image data is input to the image memory is provided, and when the trigger means generates an output, Since the image data recognition processing is started, the waiting time for image input can be reduced and high-speed processing can be performed.

[0039]

FIG. 2 is a diagram showing the configuration of a system according to an embodiment of the present invention. In the figure, reference numeral 11 denotes an object such as a printed board which is an object of alignment, and the printed board is usually provided with a cross mark for alignment. 1
Reference numeral 2 is a camera for capturing the target image of the target, 13 is an A / D converter for converting the captured image data into a digital signal, 14 is an image memory for storing the input image data, and 15 is a processing system. The processing system extracts candidate points for positioning from the image data stored in the image memory 14, performs rough positioning from each of the extracted candidate points, and then performs finer positioning on the roughly positioned candidate points. Make a match. Further, the processing system 15 stores a reference image that performs pattern matching with the target image.

FIGS. 3 and 4 are main flows of this embodiment, FIGS. 5 to 10 are detailed flows of each part of the main flow, and FIG.
25 to 25 are views for explaining the processing in this embodiment, and this embodiment will be explained with reference to this figure. (1) Overall Processing In this embodiment, the recognition processing uses the following three-step detection method. First, a plurality of position candidate points are extracted in advance from the one-dimensional luminance projection distribution of the target image in the XY directions.

That is, as shown in FIG. 11, when the recognition target is a cross shape and the brightness value of the cross mark is higher than that of the background, the one-dimensional brightness distribution of the target image has a shape in which the cross part is projected. , By detecting the position, the alignment candidate point is detected. This is the same when the luminance value of the cross mark is lower than that of the background or when the shape of the recognition target has changed, and coarse registration is performed using the features of the one-dimensional luminance projection distribution of the target image to determine the candidate points. To detect. For each candidate point, the light and shade pattern matching is performed to obtain the rough position.

That is, when a plurality of rough position candidate points are found, as shown in FIG. 12, pattern matching with the reference image is performed centering on each of the candidate points, and the candidate point with the highest correlation is finally roughed. It is the position point. As a result, the points that cannot be narrowed down by the above-mentioned projection distribution can be narrowed down to one point by this method. Then, as shown in FIG. 13, light and shade pattern matching is performed only in the extracted rough position vicinity region.

In the flow charts of FIGS. 3 and 4, steps S1 to S3 correspond to the above processing, and step S
4 corresponds to the above processing, and steps S5 to SS15 correspond to the above processing. Next, the overall processing of this embodiment will be described with reference to FIGS. Figure 3
In step S1, the reference image data is read from the image data file prepared in advance, stored in the array, and the initial value of the maximum correlation coefficient is set, as will be described later with reference to FIG.

In step S2, a value such as the self-dispersion of the reference image, which is required when the correlation value is calculated later, is calculated and read (data that is calculated in advance is read, or it is calculated and read each time). In step S3, as will be described later, the processing shown in FIGS. 6 and 7 is performed for each of the XY directions to extract candidate points that are considered to be rough positions.

In step S4, as will be described later with reference to FIGS. 8 to 10, the peak point (coordinate) having the highest correlation.
Is detected as a rough position. In step S5, it is confirmed whether or not a certain range in which the pattern matching is performed with the coarse position as the center is out of the target image. Then, when the image is out of the target image and cannot be searched, the process proceeds to step S14, an error message is output, and the process ends.

In steps S6 to S11,
Pattern matching (correlation value calculation, the calculation of correlation value will be described later) is performed at the center of the search designated area (coarse position), and the standard pattern is moved to the next pattern matching position. Similarly, pattern matching is performed. Here, the pattern matching is not a normal raster scan method, but steps S6 to S1.
As shown in FIG. 1 and A of FIG. 4, the spiral scanning is performed from the center toward the outside.

In most cases, the position to be extracted is near the center of the target image (the target image here is a part of the target image whose region is designated from the rough position).
Instead of the raster scan method (scanning in the horizontal direction, shifting to the vertical direction when reaching the end and repeating scanning in the horizontal direction) as shown in (a), as shown in FIG. A spiral scan is performed from the center of the target image. By thus spirally scanning, the position can be extracted at high speed.

In step S12, it is determined whether or not the initial value of the maximum value of the correlation coefficient has been rewritten. If it has been rewritten, the recognition position is output, and if it has not been rewritten, in step S15. Print an error message and exit. (2) Read of Reference Image Data FIG. 5 is a diagram showing a detailed flow of the read process of the reference image data in step S1 of FIG. 3, and the read process of the reference image data of this embodiment will be described with reference to FIG.

In step S1 of FIG. 5, the reference image data file provided in the overall control means 17 is opened, and in step S2, the position of the file pointer is set to the head of the file. In step S3, RGB
Type is read, the size information of the image data is read in step S4, and the size of the image is calculated in step S5.

In step S6, it is determined whether or not the previously defined value and the size match, and if they match, the brightness value data is read in step S7 and the file is closed in step S8. If the sizes do not match, the process proceeds to step S9, an error message is output, and the process ends.

In consideration of the fact that the image of the object is a cross shape, the reference image has a cross shape that covers the cross shape of the object in this embodiment.
FIG. 15 is a diagram showing an example of the reference image of the present embodiment. In FIG. 15, M1 is a cross mark as a recognition target that is marked on the object, and M2 is a reference cross-shaped reference image.

As shown in the figure, when the shape of the object to be recognized is a cross, a cross-shaped reference pattern having a shape that covers the cross makes it possible to perform processing with higher accuracy and speed compared to a normal rectangular pattern. It will be possible. That is, with the above-mentioned cross shape, compared to the rectangular pattern surrounded by the dotted line in the figure, it is not necessary to scan the shaded portion, the calculation time for that portion is reduced, and it is not affected by noise in the shaded portion. Therefore, it becomes possible to perform positioning at higher speed and with higher accuracy.

In the above flow, an error message is output when the sizes do not match, but if the sizes do not match, the magnification of the reference image is adjusted to match the predefined value, and the magnification is adjusted. However, it is possible to output an error message when the sizes do not match. That is, as shown in FIG. 16, when there is a difference in magnification between the reference image and the target image, the size of the reference image is measured, the magnification is calculated, the reference image data is multiplied by the magnification, and the pattern data is rewritten. I do. By automatically adjusting this, the recognition can be surely performed even when the magnification is deviated. (3) Extraction of Peak Candidate Points FIG. 6 and FIG. 7 are flowcharts showing the extraction processing of peak candidate points, and the extraction processing of peak candidate points will be described with reference to FIG.

First, in step S1, with respect to the array of the target image read by the image input device 12 and stored in the image memory 13, the sum of the brightness values of each row and each column is obtained, and the projection distribution of the brightness values with respect to the XY coordinate axes is obtained. Ask for. In step S2, in order to obtain a smooth one-dimensional projection distribution, the moving average value of the sum of the brightness values is obtained, and the projection distribution is reconstructed with the obtained moving average value.

In step S3, the maximum value and the minimum value of the projection distribution are obtained. In step S4, the ratio of length calculation from the peak for detecting the mountain width is set. In other words, in the present embodiment, the detection of peak candidate points is for narrow peaks, and points that have fallen some percentage of the height of the peaks are searched for from both sides of the peaks, and the points found are searched. The distance between them is the width.

Therefore, in step S4, the ratio of what percentage of the height of the mountain is to be captured is set from the difference between the maximum value and the minimum value obtained in step S3. By the way, even if the projection distribution of the same image, as shown in FIG. 17, the projection distribution may change depending on the light amount of illumination.
In such a case, it is necessary to change the ratio from the height of the mountain according to the amount of light.

That is, when it is bright, the projecting degree of the portion to be extracted is small as shown in FIG. 7C, so it is dark when it is shown in FIG. In such a case, it is difficult to extract the one with a large protrusion degree. Therefore, as shown in FIG. 18, the ratio rate
1 is changed according to the brightness, and the product of the height h of the mountain and the ratio, rate1 · h, is obtained. Then, the width of the mountain at that height is obtained, and even if the protrusion degree is small to some extent, it is extracted as a candidate for the rough position, and it is also adapted to the change in brightness. The brightness of the target image is determined from the maximum value obtained in step S3.

Returning to FIG. 6, in step S5, only peak points of peaks of the luminance value distribution other than points that deviate from the target image during pattern matching are extracted. That is, the points finally remaining as peak candidates are combined with those in both XY directions, and are used as coordinates to perform matching with the reference image. For this reason, since it becomes impossible to match with the coordinates that are too close to the edge as the center, points that are too close to the edge are removed from the peak candidates in advance,
As shown in FIG. 20, all the peaks at the positions where matching is possible in the shaded portion shown in FIG. 19 are detected.

Thereafter, the processing from step S7 to step S11 is repeated for the number of peak candidates. In step S7, it is determined whether or not the absolute value of the peak is large to some extent. If it is large, the process goes to step S8, and if it is small, the process returns to step S6. That is, as shown in FIG. 21, the second ratio rate2 with respect to the maximum value is set in advance, and peaks having peak values equal to or higher than that value are narrowed down.

In step S8, a point on the right side of the mountain where the peak is lower than the ratio set in step S4 is searched for. If not found, the process returns to step S6. If found, the process goes to step S9 in FIG. That is, as shown in FIG. 22, in the limited range on the right side of the mountain, points below the peak of the mountain that are lower than the set value (rate1 · h) are sequentially examined from the peak value, and if that point is found, step S9 is performed. go to.

In step S9, similarly to the above, a point on the left side of the mountain where the peak is lower than the ratio set in step S4 is searched for, and if not found, the mountain is regarded as not suitable as a peak candidate and the process returns to step S6.
Examine other mountains as above. If found, go to step S10. In step S10, it is determined whether or not the found mountain width is narrower than a specified value. If it is wide, the process returns to step S6 and the above process is repeated. If it is narrow, the process goes to step S11 to store the peak candidate points in the array.

That is, as shown in FIG. 23, if the distance l (lowercase letter L) of the found point is smaller than the set value,
Record the peak of the mountain as a peak candidate point. (4) Rough registration By the above processing, the peak candidates can be narrowed down considerably, but if there are multiple points that satisfy the conditions,
Matching with the reference image is performed with the coordinate point, which is a combination of points in the XY directions, as the center, and the coordinate with the highest correlation value is used as the rough positioning result.

That is, two-dimensional pattern matching (grayscale image processing) by the modified normalized correlation shown in each equation of FIG. 24 is performed on a range (two-dimensional) of a part of the target image centered on each candidate point, The candidate point having the highest value is stored as the rough position. In the figure, Xmn is reference image data and Ymn is target image data. In the above equation, the square value r ² is used instead of the correlation coefficient r to eliminate the route calculation and improve the calculation speed on the computer.
The difference between the calculated values for each image is made larger than the value to facilitate the determination (r ² takes a value of 0 to 1 to obtain the maximum position).

The coefficients C1 and C2 are the reference image data X.
Since only mn is used and it is a value that can be calculated in advance, treating it as a constant can shorten the processing time. FIG. 8 is a flowchart showing the rough alignment processing, and the rough alignment will be described with reference to FIG. In step S1, it is determined whether there is one peak candidate point for both X and Y,
In the case of one point, the process proceeds to step S9, the selected peak point is set as a rough position in a variable, and the process ends.

When there are a plurality of peak candidate points, the procedure goes to step S2, and it is determined whether the peak candidate points are not 0 in both X and Y. If either one is 0, the procedure goes to step S10 to output an error message. finish. If both peak candidate points are not 0, the process goes to step S3, loops for the number of Y candidate points, and in step S5, as shown in FIG.
The correlation value is calculated as described later in the flowchart of FIG.
Similarly, for the X candidate points, the correlation value is calculated in step S5 by looping by the number of X candidate points.

In step S6, it is determined whether or not the initial value of the maximum correlation value given in advance has been rewritten, and if it has been rewritten, an appropriate initial value has been found, so the peak point selected in step S7 is selected. Is set to a variable and the process ends. If it has not been rewritten, an error message is output and the processing ends.

9 and 10 are flowcharts showing the correlation value calculation processing in step S5, and the correlation value calculation processing will be described with reference to FIG. 15 when the reference image is the cross mark shown in FIG. In step S1, the size of the reference image in the Y direction is looped at a constant pitch. For example, when the reference image is the cross mark shown in M2 of FIG. 15, the vertical size of the cross mark is scanned at a constant pitch by the vertical size.

In step S2, it is determined whether the loop count (scanning position) is the horizontal bar portion of the cross mark, and if it is the horizontal bar portion, the process proceeds to step S3 to calculate the coefficient of the horizontal bar portion of the cross. If it is not a horizontal bar, the process goes to step S4 to calculate the coefficient of the part other than the cross horizontal bar. That is, as shown in the equation of FIG.
Since the coefficients C1 and C2 can be obtained in advance if the reference image is determined, the coefficients C1 and C2 are obtained in advance in steps S4 and S5.

In step S5, the coefficients Sxy, Sxx and Syy for calculating the correlation value shown in FIG. 24 are calculated, and in step S6 it is determined whether the denominator values Sxx and Syy are 0. Has some abnormality in the calculation process, an error message is output in step S10 and the process ends. If not 0, the square value r ² of the correlation coefficient r is obtained in step S6.

In step S8 of FIG. 10, it is determined whether the square value r ² of the correlation coefficient r is larger than a predetermined first boundary value 1, and if it is larger in step S9, the final recognition position is determined. Output as if there is, and end. If the squared value r ² of the correlation coefficient r is not larger than the predetermined first boundary value 1, the process proceeds to step S11, where r ² is the second boundary value 2 (boundary value 1> boundary) If it is larger than the value 2), it is judged whether or not r ² is larger than the maximum value of the registered boundary values. Then, if it is larger, the maximum value of the boundary value is replaced in step S13, and the process ends. Also, step S
11. In step S12, when it is determined that the value is smaller than the second boundary value 2 or smaller than the maximum boundary value 2, the process ends.

The edge of the object should be as shown in FIG. 25 (a) when viewed from the design value. However, the image actually captured is affected by various influences in addition to scratches.
It becomes as shown in FIG. Therefore, with respect to the information of the edge portion that tends to be ambiguous, the accuracy can be improved by reducing the weight applied to the correlation value. Therefore, when calculating the correlation value, the correlation coefficient is calculated for the edge portion by lowering the weight of the self-dispersion calculation value of the target image. This makes it possible to reduce the effect of scratches or the like on the edge portion. (5) Precision Positioning The coarse position obtained as described above is subjected to precision positioning by the density pattern matching.

That is, the precision alignment is performed by performing the processing of steps S5 to S15 in FIGS. Here, as shown in FIG. 14, step S7
In the process of step 11, the matching with the reference image is performed spirally around the rough position by the procedure shown in FIGS. 9 and 10, and the position with the largest correlation value is output as the recognition position. (6) Other Embodiments FIG. 26 shows another embodiment of the present invention.
An embodiment is shown in which high-speed processing is performed by using a single image memory.

In the figure, 21 is a target image memory for storing target image data, 22 is a reference image memory for storing reference images, 23 is a processing system, 24 is a monochrome camera for inputting images, and 25 is a monochrome camera. The selector 24 switches the output of 24 to the target image memory or the reference image memory. In the figure, the processing system 23 is a selector 2
4, the reference image received by the monochrome camera 24 is stored in the reference image memory 22, and the target image received by the monochrome camera 24 is stored in the target image memory 21.
To store.

When the reference image is accessed for pattern matching, the memory of the processing system 23 directly accesses the reference image memory 21. As described above, by using the two image memories, the reference image can be processed while being stored in the memory, and high-speed processing can be realized.

FIG. 27 shows that RGB memories are provided,
An embodiment is shown in which the recognition accuracy is improved by storing the color target image in each of the RGB memories. In the figure, 31 is a memory for R (red), 32 is a memory for G (green), 33 is a memory for B (blue), 34 is a color camera, and a target image received by the color camera 34. Is a memory 3 for each R, G, B
1, 33, 34 are stored. Then, the image data stored in each memory is separately processed by the same algorithm, and the final alignment is performed by averaging the results.

Alternatively, when the difference between the processing results of the image data stored in each memory is less than a certain value, the recognition result is output assuming that the correct recognition result is obtained. That is, since the images input by the color camera are slightly different in RGB, it is possible to recognize the images with the same algorithm separately more accurately than with a single monochrome screen. It is possible to prevent erroneous recognition.

In FIG. 28, a plurality of memories for storing the target image data are provided, and the image data stored in each memory is processed by different processing methods.
An example in which recognition accuracy is improved is shown. In the figure, reference numerals 41, 42 and 43 denote first to third image memories and 2
Reference numeral 4 is a monochrome camera for inputting an image, and 25 is a selector. The same target image received by the monochrome camera 25 is switched by the selector 25, and the image memories 41 to 41 are connected.
It stores in 43.

At the time of pattern matching, the first
~ Each image stored in the third image memory 41, 42, 43 is processed by a different processing method,
A final recognition result is obtained by averaging a plurality of recognition results.
Alternatively, the recognition result is output assuming that the correct recognition result is obtained when the difference between the recognition results is within a certain value.

As the separate processing method, for example, the processing method shown in the above-mentioned embodiment and the known processing method shown in the conventional example can be used. This allows
Since the respective processing methods complement each other's weak points, they can be recognized more accurately than the processing by one method. FIG. 29 shows an embodiment in which two processing systems are provided and different processing methods are used by the two processing systems to improve recognition accuracy and processing speed.

In the figure, 51 is a main controller for controlling the whole, 52 and 53 are first and second systems, and the first and second systems are processing systems 52a and 53a and image memories 52b and 53b, respectively. Is equipped with. A selector 55 switches the output of the camera 54 and stores the received image in the image memories 52b and 53b of the first and second systems.

In this embodiment, the image memory 52b,
The target image and reference image data stored in 53b are processed in parallel by different methods by the first system 52 and the second system 53, and the final recognition result is obtained by averaging a plurality of recognition results. Alternatively, the recognition result is output assuming that the correct recognition result is obtained when the difference between the recognition results is within a certain value.

Therefore, like the embodiment shown in FIG. 28,
Since each processing method complements each other's weaknesses and can be recognized accurately, processing is performed in parallel,
The processing can be performed in a short time. It should be noted that if the result of the system that finishes the fastest is used, the speed becomes faster. FIG. 30 shows an embodiment in which one target image is divided into a plurality of parts and pattern matching is performed for each of the divided parts to speed up the process.

In the figure, 61 is a main controller, 62 to 65 are first to fourth processing systems, 66 is a memory storing target image data, 67 is a camera, and the main controller 61 is the above first to first 4 processing systems 62-65
To control the whole. In this embodiment, as shown in the figure, the target image received by the camera 67 is stored in the memory 66.
, And each unit divided into ¼ of the target image by the first to fourth processing systems 62 to 65 is processed in parallel.

That is, the position of the object is recognized for each of the divided images, and the first to fourth processing systems 62 to 6 are used.
Process 5 in parallel. As a result, the processing can be performed in a processing time that is 1/4 of the normal processing time. FIG. 31 shows an embodiment in which image input and image processing can be performed in parallel to enable high-speed processing.

In the figure, (a) shows a normal processing system, and (b) shows a processing system of this embodiment. In FIGS. 1A and 1B, 71 is an image input device,
72, A for converting the output of the image input device into a digital signal
A / D converter, 73 is a processing system, and 74 is an image memory for storing image data input by the image input device.

In the same figure, in the case of normal processing, as shown in FIG. 9A, 1 to n input from the image input device 71 and converted into digital data by the A / D converter 72. After the image data of all lines (for all screens) is saved in the image memory, a trigger signal for ending one screen is generated and image processing is performed. On the other hand, in the present embodiment, as shown in FIG. 7B, a 1-line end signal is generated every time 1-line data is saved in the image memory 74, and a necessary line of the target image is output. The processing by the processing system starts at the time when only the image is saved as digital data in the image memory.

As a result, the image input operation and the processing operation can be performed in parallel, and high-speed processing can be realized.

[0088]

As described above, according to the present invention, the following effects can be obtained. (1) 1 from the one-dimensional projection distribution of the captured image in the XY directions
Or a plurality of candidate points for alignment are extracted, and the extracted one or a plurality of candidate points are subjected to light and shade pattern matching with a reference image to obtain a rough position for alignment, and the rough position is obtained. Since the light and shade pattern matching is performed with respect to the reference image with respect to the neighboring area to perform the precise alignment, high-speed processing can be performed without lowering the positioning accuracy. Also, easy rough position candidates
The rough position can be accurately extracted.
I can. (2) By using a pattern such as a cross shape having a shape and density corresponding to the recognized pattern as the reference image, the processing speed can be increased as compared with the rectangular pattern. In addition, since it is possible to cut a portion where noise is easily added, it is possible to improve accuracy. (3) By changing the condition of rough position extraction according to the brightness of the target image due to the change in illumination, accurate recognition can be performed even when the illumination at the time of recognition changes. (4) For each of the extracted one or more candidate points, the grayscale pattern matching with the reference image is performed, and the coordinate with the highest correlation value is set as the rough position, so that misrecognition at the rough positioning time is possible. Therefore, it is possible to perform accurate rough positioning. (5) When performing pattern matching with the reference image, by performing a spiral search from the center of the area near the rough position toward the outside, the time required to reach the point to be extracted can be shortened compared to the raster scan method. The processing speed can be increased. (6) When performing pattern matching with the reference image, by calculating in advance what can be calculated and processing it as a constant, the time for calculating the correlation value can be shortened, and the processing speed is increased. it can. (7) When pattern matching with the reference image is performed, weighting around the contour of the target image data is reduced to obtain the correlation value, which enables highly accurate recognition. (8) By automatically adjusting the size of the reference image and matching the sizes of the reference image and the target image to perform pattern matching, it is possible to easily recognize objects having different magnifications. (9) Since a memory for storing the reference image, a memory for storing the target image, and a selector are provided, and the images captured from the image input device can be stored in the respective memories by the selector, the reference image is dropped to a file. No need to
You can save time and effort. Further, the processing time can be shortened by directly accessing the reference image data from the memory. (10) The target image is stored in each memory for each R, G, B data, and the image data stored in each memory of R, G, B are separately processed by the same algorithm, and the above three processes are performed. By obtaining the recognition result from the average of the results,
Accurate recognition can be performed.

Further, when the difference between the above-mentioned three processing results is within a certain range, the recognition result is output assuming that the correct result is obtained, thereby reducing the possibility of erroneous recognition and performing accurate recognition. it can. (11) By storing the target image in a plurality of memories, processing the image data stored in each memory separately by different algorithms, and obtaining a recognition result from the average of the above three processing results, each algorithm It is possible to cover the shortcomings of and to accurately recognize the position.

Further, by outputting the recognition result assuming that the correct result is obtained when the difference between the above-mentioned three processing results is within a certain range, the possibility of erroneous recognition can be reduced. (12) By inputting the same target image to multiple systems, processing each system separately with different algorithms, and obtaining recognition results from the average of the above three processing results, accurate position recognition is possible. Become.

Further, by outputting the recognition result assuming that the correct result is obtained when the difference between the above-mentioned three processing results is within a certain range, the possibility of erroneous recognition can be reduced. Furthermore, by outputting the result of the fastest processing among the above three processing results as the recognition result, the recognition processing can be performed in a short time. (13) One target image is divided into a plurality of parts, and the plurality of divided target images are processed in parallel by the above system to obtain a recognition result, so that processing can be performed in a fraction of the number of divisions. , The processing can be speeded up. (14) Providing a trigger unit that generates a trigger signal every time a predetermined amount of target image data is input to the image memory, and when the trigger unit generates an output, by starting recognition processing of the target image data, Reduce the waiting time for image input,
High-speed processing becomes possible.

[Brief description of drawings]

FIG. 1 is a principle diagram of the present invention.

FIG. 2 is a diagram showing a configuration of a system according to an embodiment of the present invention.

FIG. 3 is a diagram showing a main flow of an embodiment of the present invention.

FIG. 4 is a diagram showing a main flow of an embodiment of the present invention.

FIG. 5 is a diagram showing a detailed flow (reference image read) of the embodiment of the present invention.

FIG. 6 is a diagram showing a detailed flow (candidate point extraction) of the embodiment of the present invention.

FIG. 7 is a diagram showing a detailed flow (candidate point extraction) of the embodiment of the present invention.

FIG. 8 is a diagram showing a detailed flow (rough positioning) of the embodiment of the present invention.

FIG. 9 is a diagram showing a detailed flow (correlation value calculation) of the embodiment of the present invention.

FIG. 10 is a detailed flow of an embodiment of the present invention (correlation value calculation)
FIG.

FIG. 11 is a diagram showing a one-dimensional luminance projection distribution.

FIG. 12 is a diagram illustrating extraction of coarse position points having the highest correlation value.

FIG. 13 is a diagram illustrating precision positioning.

FIG. 14 is a diagram illustrating a method of scanning a target image.

FIG. 15 is a diagram showing an example of a reference image and a target image.

FIG. 16 is a diagram illustrating a magnification adjustment of a reference image.

FIG. 17 is a diagram illustrating a difference in one-dimensional luminance projection distribution due to the brightness of illumination.

FIG. 18 is a diagram illustrating a change in extraction level of peak candidate points depending on the brightness of illumination.

FIG. 19 is a diagram illustrating a matchable area.

FIG. 20 is a diagram showing matching possible areas on a one-dimensional luminance projection distribution.

FIG. 21 is a diagram illustrating detection of rough position candidate points.

FIG. 22 is a diagram illustrating detection of rough position candidate points.

FIG. 23 is a diagram for explaining detection of rough position candidate points.

FIG. 24 is a diagram illustrating a method of calculating a correlation value.

FIG. 25 is a diagram showing an actual one-dimensional projection distribution.

FIG. 26 is a diagram showing another embodiment in which a reference image memory and a target image memory are provided.

FIG. 27 is a diagram showing another embodiment in which an RGB memory is provided.

FIG. 28 is a diagram showing another embodiment in which a plurality of image memories are provided.

FIG. 29 is a diagram showing another embodiment in which first and second processing systems are provided.

FIG. 30 is a diagram showing another example in which a target image is divided into four and parallel processing is performed.

FIG. 31 is a diagram showing another embodiment in which processing is performed every time one line of an image is input.

FIG. 32 is a flowchart showing a conventional example.

FIG. 33 is a flowchart showing a conventional example.

[Explanation of symbols]

1 Object 2 Image Input Device 3 Image Memory 4 Candidate Point Extracting Means 5 Coarse Positioning Means 6 Precision Positioning Means 7 Overall Control Means 11 Objects 12, 54, 67 Cameras 13, 72 A / D Converter 14 Image Memory 15 Processing system 21, 66 Target image memory 22 Reference image memory 23, 52a, 53a, 62 to 65, 73 Processing system 24 Monochrome camera 25, 55 Selector 31, 32, 33 RGB memory 34 Color camera 41, 42, 43, 52b, 53b, 74 Image memory 51, 61 Main controller 52, 53 First and second
System 71 image input device

─────────────────────────────────────────────────── ─── Continuation of front page (58) Fields surveyed (Int.Cl. ⁷ , DB name) H05K 13/04 G01B 11/00 G06T 1/00 G06T 7/00 G06T 7/60

Claims (20)

(57) [Claims] 1. A target image on a member on which a mark for alignment is marked is captured, and one or a plurality of candidate points for alignment are extracted from a one-dimensional projection distribution in the XY direction of the captured image. , The extracted one or more candidate points are subjected to a light and shade pattern matching with a reference image to obtain a rough position for alignment, and a light and shade pattern matching with the reference image is performed for the rough position neighborhood region. A positioning method characterized by performing precise positioning. 2. An image input means for capturing a target image on a member on which a mark for alignment is marked, an image memory for storing the captured image data, and an image captured by accessing the image memory . XY
The candidate point extracting means for extracting one or a plurality of candidate points from the one-dimensional projected distribution in the direction ; the rough positioning means for roughly aligning the extracted one or a plurality of candidate points; and the rough positioning means. A position characterized by including a precise positioning means for performing a finer position matching by performing a light and shade pattern matching with the reference image only in the obtained rough position vicinity area, and an overall control means for controlling each of the above means. Matching device. 3. The alignment apparatus according to claim 2, wherein a pattern optimal for recognizing a target image having a shape and density corresponding to the pattern to be recognized is used as the reference image. 4. The alignment apparatus according to claim 2 , wherein the candidate point extraction means changes the rough position extraction condition according to the brightness of the target image due to the change in illumination. 5. The coarse positioning means performs a light and shade pattern matching with a reference image for each of the extracted one or a plurality of candidate points, and sets the coordinates having the highest correlation value as a rough position. Claim 2, 3 or claim 4
Alignment device. 6. precision positioning means, when performing the pattern matching with the reference image, according to claim 2, 3, 4, or claims from the center of the coarse position neighboring region outwardly, characterized in that the search in a spiral The alignment device according to item 5 . 7. When the pattern matching with the reference image, initially calculated in advance what can be calculated, according to claim 2, 3, 4, 5, characterized in that the process as a constant or
Or the alignment device according to claim 6 . 8. When performing pattern matching with a reference image, the weight around the contour of the target image data is reduced to obtain the correlation value .
The alignment device according to claim 6 or claim 7 . 9. The pattern matching is performed by automatically adjusting the size of the reference image when the sizes of the reference image and the target image are different from each other so that the sizes of the reference image and the target image match. 2,3,4,5,6,7 or contract
Positioning device according to claim 8 . 10. A memory for storing a reference image, a memory for storing a target image, and a selector are provided, and an image captured from an image input device can be stored in each memory by the selector. Terms 2, 3,
The alignment device of claim 4, 5, 6, 7, 8 or claim 9 . 11. When a leading reference image data, the alignment apparatus of claim 10, characterized in that direct access to the memory storing the reference image. 12. A memory for storing respective image data of R, G, B is provided, and a target image is inputted as a color image and stored in each of the memories for each of R, G, B data.
R, G, and treated separately by the respective same algorithm stored image data to each memory B, claims 2,3,4, characterized in that to obtain a recognition result from the average of the three processing results , 5, 6, 7, 8, 9, 10 or contract
Motomeko 1 1 of the alignment device. 13. A memory for storing respective image data of R, G, B is provided, and a target image is inputted as a color image and stored in each of the memories for each of R, G, B data.
The image data stored in the R, G, and B memories are separately processed by the same algorithm, and when the difference between the above three processing results is within a certain range, the recognition result is obtained as a correct result. A contract characterized by outputting
Demand 2,3,4,5,6,7,8,9,10 or claim
Item 11. The alignment device according to item 11 . 14. A plurality of memories for storing target image data are provided, the target images are stored in the plurality of memories, and the image data stored in each memory are separately processed by different algorithms to obtain the three images. The recognition result is obtained from the average of the processing results .
The alignment device according to claim 6, 7, 8, 9 , 10 or claim 11 . 15. A plurality of memories for storing the target image data are provided, the target images are stored in the plurality of memories, and the image data stored in each memory are separately processed by different algorithms to obtain the three images. 6. The recognition result is output as if the correct result was obtained when the difference between the processing results is within a certain range .
The alignment device according to claim 6, 7, 8, 9 , 10 or claim 11 . 16. A plurality of systems for performing recognition processing are provided, the same target image is input to the plurality of systems, each system separately processes by different algorithms, and recognition is performed based on an average of the three processing results. Claims 2, 3, 4, 5, 6, 7, 8, characterized in that a result is obtained .
The alignment device according to claim 9, 10, or 11 . 17. A plurality of systems for performing recognition processing are provided, the same target image is input to the plurality of systems, and each system separately processes by different algorithms, and the difference between the three processing results is constant. The recognition result is output as if the correct result is obtained when the value is within the range, and the recognition result is output .
10 or alignment apparatus of claim 1 1. 18. A plurality of systems for performing recognition processing are provided, the same target image is input to the plurality of systems, and each system separately processes by different algorithms, and among the three processing results, the most The result processed at high speed is output as a recognition result.
2,3,4,5,6,7,8,9,10 or claim 1
1 alignment device. 19. A plurality of systems for performing recognition processing are provided, one target image is divided into a plurality, and the plurality of divided target images are processed in parallel by the system to obtain a recognition result. Claims 2, 3, 4, 5, 6,
The alignment device according to claim 7, 8, 9, 10, or 11 . 20. A trigger means for generating a trigger signal each time a predetermined amount of target image data is input to the image memory is provided, and when the trigger means generates an output, the recognition processing of the target image data is started. 3. The method according to claim 2, wherein
3,4,5,6,7,8,9,10,11,12,1
The alignment device according to claim 3, 14, 15, 16 , 17 , or 18 or claim 19 .