JPH04233402A - Position recognizing system of object using tv camera - Google Patents

Abstract

PURPOSE:To shorten processing time of a processor while ensuring accuracy of position recognition by setting a detection area from positional information of a binary data to determine a correct position of an object from multi-contrast data of pixels in the area CONSTITUTION:Object 1 is taken with TV camera 2 and a multi-contrast data and a binary data are determined from an image data to be stored into a variable density image memory 3 and a binary image memory 4 respectively. The memory 4 is divided into block areas comprising pixels and a scanning is performed at each block area 5 to determine a position coordinates of position in the memory 4 of the pixel which presents 1 first. A detection area larger comparatively than the whole of an object is set on the memory 3 based on the coordinates. A correlation value of the multi-contrast data is determined between the pixels of a template 8 of an object 1 and the corresponding pixels in the memory 3 in the area 7. The position of the template 8 given when the correlation value is the maximum is determined as the position of the object 1. In this manner, a processor 9 performs a computation linearly for the pixels in the area 7 based on the multi-contrast data thereby enabling the shortening of processing time with the processor 9.

Description

[Detailed description of the invention]

0001

[Field of Industrial Application] The present invention relates to a method for recognizing the position of an object using a television camera, and in particular to multi-gradation data and binary data for photographed images of objects such as IC packages and alignment patterns. First, the rough position of the object is recognized based on the binary data, and the detection area based on this position information is stored in a memory storing multi-tone data. By setting the detection area in This invention relates to a method for recognizing the position of an object.

0002

2. Description of the Related Art Conventionally, as shown in FIG. 7, as a method for recognizing the position of an object using a television camera, a multi-gradation image obtained by photographing an object 71 such as a positioning pattern with a television camera 72 is used. The data is stored in the grayscale image memory 73, and a window 74 of a size corresponding to M×N pixels (M and N are positive integers) that can completely capture the object 71 is created in the grayscale image memory 73. The processor 76 calculates the correlation value between each pixel of the template 75 having the same size as the window 74 and the corresponding pixel of the original image in the window 74. Through arithmetic processing, the template 75 is shifted pixel by pixel and determined at each shift position, and the position of the template 75 at which this correlation value is maximum is determined to be the position of the object. The method is common.

[0003] The template 75 used here is
This is a template created from multi-gradation data of each pixel in the object portion and background portion, which is determined based on the design value of the object 71 to be detected, taking into account variations.

0004

As described above, in the conventional method, each pixel area of the template 75 is changed every time the template 75 is shifted, even for a pixel area in the grayscale pixel memory 73 where the object 71 does not exist. The processor 76 calculates the correlation value between the multi-tone data and each corresponding pixel in the grayscale image memory 73. Therefore, although the accuracy of position confirmation is sufficient, there is a problem that calculations in the processor 76 take time.

Therefore, in the present invention, a binary image memory is prepared in addition to the grayscale image memory, and each memory stores multi-gradation data and binary data regarding the image of the object. First, the rough position of the object is recognized based on the binary data in the binary image memory, and the detection area based on this position information is set in the grayscale image memory storing multi-tone data. , and then determine the exact position of the object based on the multi-tone data of each pixel in this detection area, reducing the processing time required by the processor and increasing the accuracy of position recognition. The purpose is to ensure that

[0006]

[Means for Solving the Problems] The present invention stores multi-gradation data and binary data of an image obtained by photographing an object in separate memories, and first the processor The rough position of the object is recognized using the value data, and then the multi-tone data of each pixel in this recognized range is
The position of the object is determined by determining the correlation value between each pixel of a template prepared in advance and multi-gradation data.

FIG. 1 is a diagram explaining the principle of the present invention. In the figure, reference numeral 1 indicates an object, for example, a cross-shaped pattern for positioning. 2 is a television camera;
An object 1 is photographed and a grayscale image thereof is obtained. 3 is
This is a grayscale image memory, and stores multi-tone data obtained by A/D converting the grayscale image of the object 1. 4 is a binary image memory, in which binary data obtained by binarizing multi-tone data using a predetermined threshold value, for example, pixels of an object part are
”, data with pixels in the background portion set to “0” is stored. Note that this binarization may be performed on the output signal of the television camera 2, or may be performed on the multi-gradation data once stored in the memory 3. Furthermore, the binary image memory in the present invention refers to a memory that ultimately stores binary data. 5 is a block area,
Each block area consists of a plurality of pixels. 6 is one of the position coordinates, and for each block area 5, for example, the x coordinate and y coordinate in the binary image memory 4 of the pixel that becomes "1" when scanning from the upper left pixel in the direction of the arrow. expressed in coordinates. 7 is a detection area, and the grayscale image memory 3 is stored based on the position coordinates 6 in each of the block areas 5 including the pixels of the object part.
This is an area larger than the entire object. 8
is a template, which is created from multi-gradation data of each pixel of the object part and the background part based on the design values of the object 1. A processor 9 performs arithmetic processing to obtain correlation values, position coordinates 6, etc. for multi-tone data between each pixel of the template 8 and each pixel of the grayscale image memory 3.

Here, the procedure for recognizing the position of the object 1 based on the image taken by the television camera 2 is as follows.

That is, (1) the object 1 is photographed by the television camera 2 and image data thereof is obtained; ■ Obtain multi-gradation data and binary data from this image data, and store the multi-gradation data in the gradation image memory 3 and the binary data in the binary image memory 4. do. (2) Divide the binary image memory 4 into block areas 5 each consisting of a plurality of pixels. ■For each block area,
For example, the x and y coordinates in the binary image memory 4 of the pixel that first becomes "1" when scanned in the direction of the arrow from the upper left pixel are determined. (2) Based on the position coordinates, a detection area 7 larger than the entire object is set on the grayscale image memory 3. ■ In this detection area 7, the template 8 of the object 1 is shifted one pixel at a time, and the multi-gradation between each pixel of the template 8 at each shift position and each corresponding pixel in the grayscale image memory 3 is Find the correlation value between data. ■Template 8 when this correlation value is maximum
The position is determined to be the position of object 1. Through these steps, the position of the object 1 is recognized.

0010

In the present invention, first, the rough position of the object 1 is determined in the binary image memory 4, and the detection area 7 is set in the grayscale image memory 3 based on this position information. Then, using the conventional method for this detection area 7, that is, shifting the template 8 pixel by pixel, and calculating the statistics between each pixel of the template 8 at each shift position and the corresponding pixel in the detection area 7. A method is applied in which the processor 9 determines the target matching state.

In this manner, the processor 9 scans the binary image memory 4 to set the detection area 7 including the entire object 1, and converts only the pixels of this detection area 7 into multi-tone data. Since the arithmetic processing based on the above information is executed, the processing time in the processor 9 is shortened.

0012

[Embodiment] An embodiment of the present invention will be explained using FIGS. 2, 3, 4, 5, 6, and 7.

FIG. 2 is an explanatory diagram showing an outline of an apparatus for recognizing the position of components mounted on a printed board. In Fig. 2, 21 is a printed board, 22 is a mark indicating the component mounting position on the printed board, and 23 is an X on which the printed board is placed.
- Y table, 24 is a component mounted on the printed board,
25 is, for example, a cross-shaped pattern provided on the part for position recognition, 26 is a television camera, 27 is an image processing unit, 28
is a drive control section.

The image processing section 27 also includes a television camera 2.
A/
A D converter 31, a binary image memory 32 that stores this multi-gradation data once and finally converts it into binary data and stores it; Grayscale image memory 33 for copying and storing data, ROM 34 for storing various data, ROM 3 for storing programs
5. Processor 36 capable of high-speed calculation processing, RAM 37 and CPU 3 for holding data in the middle of calculation
It consists of 8 mag.

The drive control unit 28 also includes a controller 41 that performs control based on a control signal corresponding to the position of the pattern 25, a component clamper 42 that is controlled by this controller 41, and a pulse motor 43 that is also controlled by the controller 41. , and an XY table drive section 44 driven by this pulse motor 43.

Here, the printed board 21 and the component 24 are aligned as follows. (1) First, multi-gradation data is binarized for data compression. That is, the number of images corresponding to each density in the binary image memory 32 storing, for example, 256 levels of multi-tone data obtained by A/D converting the output signal of the television camera 26 is counted. Then, a density histogram as shown in FIG. 3 is created. Since the number of pixels of the entire pattern 25 is known in advance based on the design value, the number of pixels is accumulated in order from the brightest side of the density histogram to find the density that is closest to this known number of pixels. The multi-gradation data is binarized using the density as a threshold value, and this binary data is stored in the binary image memory 32. Note that this density histogram is created with the pattern 25 portion as “bright” and the background portion as “dark”.

(2) Next, the rough position of the pattern 25 is recognized based on the binary data in the binary image memory 32. That is, the binary image memory 32 is divided into equal parts as shown in FIG. 4 by the size of the block area 51 that is predetermined based on the size of the pattern 25 (maximum length in the x and y directions). Then, each block area is sequentially scanned in a predetermined direction, for example, from the upper left and lower right, in the direction of the arrow for each line and pixel, with the scanning order of the block areas set in the x direction, until a pixel of "1" is found. If so, the scanning within that block area is stopped at that point and the x, y coordinates of the pixel in the binary image memory 32 are determined. Then, x, y obtained for each block area scanned so far
Maximum and minimum values in the coordinates xmax, xmin
, ymax, ymin are determined each time the scanning within each block area is completed. Here, in order to shorten the time required for scanning, scanning may be performed, for example, every other line or every other pixel in each block area.

Note that scanning is terminated in the following cases. - When all block areas have been scanned. - When the value of (ymax - ymin) is approximately equal to the length in the y direction of the pattern 25 that is the object, and scanning of the block area three blocks below including the block area where the pixel ymin is located has been completed. This is because the length of the pattern 25 in the y direction is such that it fits within three continuous block areas in the y direction. Also, in the latter case, (x
Pattern 2 where the value of max − xmin ) is the object
If the scanning of the block area three blocks to the right including the block area that is approximately equal to the x-direction length of 5 and has xmin pixels is completed, the x-direction range of the block area to be scanned for the same reason. shorten. Regarding the end of scanning when the order of scanning between block areas is set in the y direction, x and y are interchanged in the above explanation.

(3) Next, xmax, x obtained earlier
Based on the values of min, ymax, and ymin, a larger detection area 52 as shown in FIG. 5 is set in the grayscale image memory 33, which has a large amount of information.

(4) Next, the exact position of the pattern 25 is determined based on the multi-gradation data of each pixel in the detection area 52. That is, pattern 2 as shown in FIG.
A window with a size equivalent to M×N pixels (M and N are positive integers) that can completely capture the pattern 25 and the multi-gradation of each pixel of the pattern 25 and its background part with the same size as this window. A template 53 consisting of data is determined in advance, and this template 53 is shifted from the upper left pixel of the detection area 52 one pixel at a time in the x direction as shown in the figure, and the template 53 at each shift position is , a correlation value with the original image in the grayscale image memory 33 is obtained through arithmetic processing in the processor 36.

[0021] Then, this correlation value is
Let Sxx be the self-variance, which is an index representing the variation from the average density of , Syy be the self-variance of the original image, and Sxy be the covariance between the template 53 and the original image,

It is obtained by [Equation 2]. Here, −1≦r≦1, and the closer it is to “1”, the greater the degree of overlap between the template 53 and the original image.

[0022] Furthermore, Sxx is defined as multi-gradation data of each pixel of the template 53 as xmn.

[Equation 3] Similarly, Syy takes the multi-gradation data of each pixel of the original image in the window as ymn,

[Math 4] is required. Also, Sxy is

It is obtained by [Equation 5]. The correlation value r between the template 53 and the corresponding original image in the grayscale image memory 33 is

[Math 6] is required.

Furthermore, in the present invention, if the template 53 is fixed, the

Considering that the part of [Formula 7] is a constant and that the correlation value r is a decimal point smaller than 1, rounding errors may occur in the arithmetic processing in the processor 36.

[Equation 8] However, C1: 1000 M: Size of the template 53 in the x direction N: Size of the template 53 in the y direction The value of multi-gradation data of each pixel is calculated and used as a correlation value. In addition, C1
, C2 and C3 are calculated in advance and used as RO for data.
Store it in M34.

Therefore, 0≦r2≦1000, and the shift position of the template 53 where r2 becomes maximum is determined as the position of the pattern 25. Note that this formula ■' cannot recognize the case where the original value of r is negative, so before determining the position of pattern 25, the numerator of formula ■' is

[Math. 9] We need to make sure that is not a negative value.

[0025]

According to the present invention, a grayscale image memory and a binary image memory are prepared, and each memory stores multi-gradation data and binary data regarding an image of an object. First, the rough position of the object is recognized based on the binary data in the binary image memory, and then the detection area based on this positional information is stored as multi-tone data. Set it in the gray scale image memory, then shift the template of the object pixel by pixel in this detection area,
The processor calculates the correlation value between the template and the corresponding original image at each shift position, and the position of the template at which this correlation value is maximum is used to determine the position of the target object. Multi-tone data in a pixel area where
The calculation for the data can be omitted, the processing time in the processor can be shortened, and the accuracy of position determination can be ensured.

Furthermore, in the calculation process of the correlation value, for example, when calculating the self-variance Sxx, it is necessary to first transform the above-mentioned formula ``■'' into the formula ``■'', so that it is not necessary to separately calculate only the average value. Executes arithmetic processing, and also 0≦
Since the correlation value r is modified so that r2≦1000, it is possible to reduce the processing load on the processor and to suppress the occurrence of rounding errors and the like during the arithmetic processing process.

[Brief explanation of the drawing]

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

FIG. 2 is an explanatory diagram showing an outline of an apparatus for recognizing the position of components mounted on a printed board according to the present invention.

FIG. 3 is a density histogram showing each density of multi-gradation data and the number of pixels having this density.

FIG. 4 is an explanatory diagram showing a state in which a binary image memory is equally divided into block areas each consisting of a plurality of pixels.

[Fig. 5] Position coordinates of the object xmax, xmin,
It is an explanatory diagram showing a detection area set based on ymax and ymin.

FIG. 6 is an explanatory diagram showing the state of a window and a template in a detection area.

FIG. 7 is an explanatory diagram showing an overview of a conventional object position recognition method.

[Explanation of symbols]

In Figure 1, 1...Object 2...TV camera 3... Grayscale image memory 4... Binary image memory 5...Block area 6...Position coordinates 7...Detection area 8...Template 9...processor

Claims (4)

[Claims] [Claim 1] Image data obtained by photographing an object with a television camera is stored in a memory, and a processor performs arithmetic processing on the image data in the memory to recognize the position of the object. In the system, the memory includes a grayscale image memory and a binary image memory, the grayscale image memory stores multi-tone data of the image, and the binary image memory stores two-value data of the image. By storing value data, dividing the binary image memory into block areas each consisting of a plurality of pixels, and scanning each of the block areas within a predetermined range, the binary data of the object is obtained. The positional coordinates of a specific one among the existing pixels are determined for each block area, and based on the positional coordinates, a detection area larger than the entire object is set on the grayscale image memory, and in this detection area, A template created in advance based on the multi-gradation data of each pixel of the object is shifted pixel by pixel, and each pixel of the template at each shift position is compared with each corresponding pixel in the grayscale image memory. The position of the template at which the correlation value becomes maximum is determined to be the position of the object, and first, the position of the template is determined to be the position of the object.
The position of the object by the television camera is characterized in that the approximate position of the object is recognized based on the data, and then the exact position of the object is determined based on the multi-tone data. Recognition method. 2. The threshold value when obtaining binary data is determined by creating a density histogram showing each density of multi-tone data and the number of pixels having this density, and then calculating the brightness and darkness of this density histogram. Accumulate the number of pixels from that density,
2. The method for recognizing the position of an object using a television camera according to claim 1, wherein the cumulative value is the density closest to a previously known number of pixels of the object. 3. The positional coordinates are the x coordinates and 2. The method for recognizing the position of an object using a television camera according to claim 1, wherein the y-coordinate is the y-coordinate. 4. The correlation value is: [Formula 1] where: r: Correlation value c1: Constant M: Size of the template in the x direction N: Size of the template in the y direction xmn: Multi-gradation data of each pixel of the template 2. A method for recognizing the position of an object using a television camera according to claim 1, wherein: time is determined by multi-tone data of each corresponding pixel in a grayscale image memory.