JPH0769959B2 - Vertex detection method for pattern recognition - Google Patents

Description

TECHNICAL FIELD The present invention relates to a pattern recognition technique that can be effectively used when measuring the position of a member or the like in material handling and the like accompanying automation of a processing machine. In particular, it relates to a method for detecting vertices for pattern recognition.

[Prior Art] The inventors of the present invention have previously proposed a method of performing pattern recognition or the like by detecting each vertex of a graphic pattern in Japanese Patent Publication No. 56-47581.

It is a proposal for a method called apex detection by contour tracking in the present specification. Specifically, a target object is captured as a two-dimensional figure pattern by a pattern input device such as a television camera, and each of the scan lines is scanned. The output of the sample points is converted into a binary signal according to the brightness of the sample points by AD conversion, and the binarized matrix pattern thus obtained is subjected to the following processing to determine the shape of the figure pattern, etc. To recognize.

In the processing of the binarized matrix pattern, attention is paid to adjacent 2 × 2 mesh points in the binarized matrix pattern, and any data at these four points is binarized 0 or 1 Depending on whether the contour of the graphic pattern, that is, the data at the boundary line between the real part and the empty part, is converted into a signal for marking the direction of the line segment of the contour line, and then converted. The pattern contour portion in the conversion information is circulated once, and the existence of the apex (corner) of the pattern is detected based on the conversion information, and the data is used for recognizing the graphic pattern.

When a figure pattern is recognized by such a method, as described above, the existence of vertices is detected by making a round of the pattern contour portion in the conversion information. Inevitably, the apex detection algorithm is also premised on detecting the existence of the apex by going around the pattern contour portion. Therefore, when the matrix pattern does not capture the whole image of the figure pattern, that is, when the figure pattern reaches the peripheral portion of the matrix pattern, the vertex detection cannot be performed and it is possible to do so. In order to do so, it is necessary to enable vertex detection by a separate algorithm.

[Problems to be Solved by the Invention] An object of the present invention is to provide a case in which the entire image of a figure pattern is not captured in the above-mentioned matrix pattern, and therefore the figure pattern reaches the peripheral portion of the matrix pattern. Also, the general vertex detection algorithm, that is, the vertex detection algorithm used when the whole image of the figure pattern is captured in the matrix pattern can be used as it is, and the processing device can be simplified. .

[Means for Solving the Problems] The vertex detection method of the present invention for achieving the above object captures a target object as a two-dimensional graphic pattern, and outputs the output for each sample point according to the brightness of the sample point. Data in the contour portion of the graphic pattern in the vertex detection processing for all or a part of the binarized matrix pattern obtained by
Depending on whether the data of the adjacent mesh points is 0 or 1 which is binarized, it is classified with respect to the direction of the contour line, and the classified data is traced along the contour line of the graphic pattern to obtain the graphic pattern. In the pattern recognition process for detecting the vertices of the pattern, a blank part with the number of pixels necessary for classifying the data along the outline of the figure pattern is attached around the detection process area in the matrix pattern. , By enlarging the detection processing area and performing apex detection processing for one round along the contour portion of the figure pattern for the expanded detection processing area,
It is characterized by detecting the vertices of the figure pattern, and further, after performing the above-mentioned vertex detection processing, among the vertices of the detected figure pattern, the vertices detected at the position in contact with the enlarged portion of the detection area are detected. It is characterized in that the others are detected as the vertices of the figure pattern.

[Examples] Examples of the present invention will be described in detail below.

Generally, in pattern recognition for measuring the shape, position, etc. of an object placed in a measurement area, the target object is captured as a two-dimensional figure pattern by a pattern input device such as a television camera, and each sample on the scanning line is captured. The output of the points is converted into a binary signal according to the lightness and darkness of the sample points by AD conversion, and the shape and position of the figure pattern is recognized by the processing of the binary matrix pattern obtained by this, and the recognition There are various methods of processing. In these methods, usually, the vertices in the contour portion of the binarized graphic pattern are extracted, and the graphic pattern is recognized based on the data.

When recognizing such a graphic pattern, as described above, it is necessary to capture the whole image of the graphic pattern in the matrix pattern in order to detect the existence of vertices by going around the outline portion of the graphic pattern. The vertex detection algorithm based on the above is adopted, but this point is the same in the present invention.

However, such an apex detection algorithm should be used as it is when the whole image of the figure pattern is not captured in the matrix pattern, that is, when the figure pattern reaches the peripheral portion of the matrix pattern. I can't.

To deal with such a problem, in the method of the present invention, by performing the processing described later on the matrix pattern in which the entire image of the graphic pattern is not captured,
It enables the vertex detection by the above algorithm.

The present invention can be applied to the case of recognizing the shape and position of a graphic pattern by processing a binarized matrix pattern by various algorithms. Further, the present invention can be applied to all or a part of the binarized matrix pattern. It can also be applied to the case of performing vertex detection processing, or when the figure pattern reaches the peripheral portion of the matrix pattern for some reason, for example, when only a part of the figure pattern is captured by the pattern input device. However, here, in the attitude angle / position measuring method of the figure pattern by the affine inverse transformation proposed by the present inventors in Japanese Patent Application No. 62-256804, the position of the feature point corner is detected with high accuracy. When a window is set and the inside of this window is used as the vertex detection processing area (proposed by the application on the same day: Position measurement of the feature point corner of the eye method ") will be described.

First, an outline of a posture angle / position measurement method using the above-mentioned affine inverse transformation will be described. In the measurement method, a graphic pattern to be measured is a matrix pattern binarized by a pattern input device such as a television camera. It is input to the image processing device. In this case, in order to improve the recognition accuracy of the graphic pattern, a relatively large capacity memory of about 1024 × 1024 is used.

In the image processing device, paying attention to adjacent 2 × 2 mesh points in the matrix pattern,
Depending on which data at those four points is a binary 0 or 1, the contour of the graphic pattern, that is, the data at the boundary between the real part and the empty part,
The vertices of the figure pattern are detected by classifying with respect to the direction of the contour line and tracing the classified data along the contour line of the figure pattern. The coordinates are calculated.

On the other hand, the reference pattern in the case where the figure pattern is at the reference position is stored in advance in the storage device in the image processing apparatus together with other various reference patterns, and the figure pattern subjected to the above-mentioned vertex detection is subjected to affine inverse transformation. Prior to measuring the posture angle / positional deviation with respect to the reference pattern, which reference pattern has a correspondence relationship, and which specific vertex in the figure pattern corresponds to which data group of the corresponding reference pattern vertex Is determined.

When measuring the posture angle / positional deviation by the affine inverse transformation between the graphic pattern and the reference pattern, the specific vertex with the largest mutual distance between the two detected by contour tracing is set as the feature point corner, and The posture angle and the positional deviation of the graphic pattern are obtained from the coordinate relationship with the feature point corner on the corresponding reference pattern.

The posture angle of the graphic pattern previously proposed by the present inventors
In the position measurement method, the figure pattern is recognized through the above process. In this case, if a relatively large capacity memory of about 1024 × 1024 is used to improve the recognition accuracy, the data processing can be performed. It takes a very long time.

In order to reduce the data processing time, a "window" as described below is applied to the binarized matrix pattern, and a part of the binarized matrix pattern is used as a vertex detection area. As a result, the data processing time can be shortened. In this case, it is premised that the schematic position of the feature point corner is known in advance for setting the window, but such a schematic position is determined by the pattern input device such as a television camera and the recognition target object. Is known without needing to measure it when the positional relationship between the target object and the target object is always within a certain range, and a simple processing of the binarized matrix pattern of the target object (the inventors of the present application filed on the same day and separately proposed. It can be obtained easily and in real time by the title of the invention “method of measuring outline shape of figure pattern”).

When the approximate position of the feature point corner is known, a window is set around the feature point corner, and data processing only within the window enables accurate coordinate measurement of the feature point corner in a short time. become. The horizontal and vertical sizes of the windows need to be set appropriately in relation to the coordinate values of the vertices adjacent to the feature point corners.

When the feature point corner is recognized by such a method, a part of the figure pattern inevitably reaches the window, that is, the peripheral part of the detection processing area in the matrix pattern. Therefore, even if an attempt is made to detect the existence of vertices by making a round in the pattern contour portion, it becomes impossible to make a round by reaching the periphery of the window, and it is not possible to use a general vertex detection algorithm for contour tracing.

In order to deal with such a problem and to make available a general vertex-tracking vertex detection algorithm, prior to the vertex detection in the matrix pattern, the figure pattern is formed around the entire window of the matrix pattern. A blank portion having a width of several pixels necessary for the vertex detection processing, which makes one round along the contour portion, is provided, thereby expanding the detection processing area and performing vertex detection by contour tracing by the above algorithm. The provision of the blank portion is a process performed for the four sides around the recognition area regardless of whether the graphic pattern is outside the recognition area.

When attaching the blank portion, first, in the MN coordinate system of the image memory corresponding to the window as schematically shown in FIG. 1, the m × n matrix of FIG.
Assuming that there is a contour portion of the figure pattern as shown in the pattern, and further that a blank portion of 2 pixels is required in the surroundings in the vertex detection algorithm by contour tracing, the binarized matrix pattern is as shown in FIG. As shown in (B), it is enlarged to (m + 4) × (n + 4), and the contour portion of the graphic pattern at the position of the dotted line is moved to the position of the solid line.

The apex detection algorithm by contour tracing, which is performed after this, generally assumes that the whole image of the figure pattern is captured in the matrix pattern in order to detect the existence of the vertices by going around the contour part of the figure pattern. Therefore, the expansion width of the matrix pattern is
It is set to a required value in this vertex detection algorithm. Specifically, as described above, attention is paid to adjacent 2 × 2 mesh points in the binarized matrix pattern, and any data at these four points is a binarized 0 or 1. When the signal conversion in the contour portion of the graphic pattern is performed by the above, an enlarged portion of 2 pixels is provided in each periphery of the m × n matrix pattern.

As shown in the flowchart of FIG. 2, the expansion of the detection processing area is performed by reducing the values of i and j from m and n, respectively, while the data of the image address (i, j) is changed to the image address (i + 2). , j + 2), and finally clear the data in the surrounding enlarged portion to complete.

When the detection processing area is enlarged by adding a blank portion in this way, the image memory corresponding to the window and the image memory for copying having the capacity including the enlarged portion are not separately prepared, and the window in the image memory is not prepared. The same area can be used, and the memory usage efficiency becomes very high. Moreover, since the operation principle is very simple and the processing is simple, it can be easily implemented as hardware and high-speed processing is possible. By performing this processing in advance, a complicated judgment processing function of whether or not the figure pattern has reached the edge of the recognition area during contour tracing, that is, an area edge detection algorithm is added to increase the processing amount and processing time. Without using the conventional simple contour tracing algorithm, contour tracing can be performed at high speed.

Next, with respect to the detection processing area enlarged as described above, the vertex detection processing for making a round along the contour portion of the graphic pattern, that is, the vertex detection by the contour tracing described above is performed.
Because of the enlargement, it is possible to make a round of the contour portion to convert the signal, etc., so that the generally used vertex detection algorithm can be used as it is.

Further, when the vertices are detected by such a method, the vertices that should not be originally detected are detected at the position in contact with the enlarged portion of the detection area.
Therefore, it is necessary to extract the original feature point corners excluding them, but this processing is performed by expanding the detection area if the coordinate values of the vertices detected by contour tracing satisfy any of the following inequalities. If it is a generated vertex, it may be excluded from the feature point corner.

Mi ≦ 3, Mi ≧ m + 2, Ni ≦ 3, Ni ≧ n + 2 Note that the calculation of such inequalities is a simple integer calculation, and whether or not the detected vertex is the feature point corner to be obtained. Since it can be determined whether or not it can be processed at a very high speed.

[Effects of the Invention] According to the method of the present invention described in detail above, when the entire image of the graphic pattern is not captured in the matrix pattern, and therefore the graphic pattern reaches the peripheral portion of the matrix pattern. Also, a general vertex detection algorithm can be used as it is, and the processing device can be simplified.

[Brief description of drawings]

1 (A) and (B) are explanatory views of the method according to the present invention, and FIG. 2 is a flow chart for explaining the expansion of the detection processing area in the above method.

Claims (2)

[Claims] 1. A target object is captured as a two-dimensional graphic pattern, and the output for each sample point is converted into a binarized signal corresponding to the lightness and darkness of the sample point, and a binarized matrix pattern obtained by this is obtained. At the time of the vertex detection processing for all or part of the area, the data in the contour portion of the graphic pattern is
Pattern recognition for classifying the direction of the contour line depending on whether the value is 0 or 1 and tracing the classified data along the contour line of the figure pattern to detect the vertices of the figure pattern. In the processing, the detection processing area is enlarged by adding a blank area around the detection processing area in the matrix pattern, which has the number of pixels required to classify the data along the outline of the graphic pattern. A vertex detection method for pattern recognition, characterized in that the apex of the figure pattern is detected by performing apex detection processing that makes a round along the contour portion of the figure pattern for the enlarged detection processing area. . 2. A target object is captured as a two-dimensional graphic pattern, and the output for each sample point is converted into a binarized signal according to the lightness and darkness of the sample point, and the binarized matrix pattern obtained thereby is obtained. In the vertex detection processing for all or some areas, the data in the contour portion of the graphic pattern is classified with respect to the direction of the contour line depending on whether the data of the adjacent mess point is binary 0 or 1. In the pattern recognition processing of tracing the classified data along the contour line of the figure pattern and detecting the vertices of the figure pattern, the contour part of the figure pattern is formed around the detection processing area in the matrix pattern. A blank area with the number of pixels required to classify the above data is added along the line to expand the detection processing area. For the enlarged detection processing area, the vertex detection processing is performed along the contour portion of the figure pattern, and the detected vertex is detected at a position in contact with the enlarged portion of the detection area. A vertex detection method for pattern recognition, which is characterized by detecting vertices and detecting the others as vertices of a figure pattern.