JP3126491B2 - Pattern recognition method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a pattern recognition method for recognizing a target pattern obtained by capturing an image using a video camera.

[0002]

2. Description of the Related Art Conventionally, a method of comparing a target pattern obtained by imaging using a video camera with a predetermined reference pattern and detecting a defect of the target pattern based on whether or not these patterns match. Has been proposed (JP-A-60-154104).

There is a method of recognizing an embossed character such as a car body number imprinted on a predetermined portion such as a cowl panel of a car body using this method. That is, in this embossed character recognition method, the embossed character is imaged by using a video camera, and the obtained image (target pattern) is, for example, vertically and horizontally.
It is divided into 25 sections (areas), and the identification value of each section is classified into three levels according to the gray level in each section. In addition, the target pattern is divided into 25 sections, and a reference pattern in which a judgment value is preset in each section is stored in advance, and the identification value of each section of the classified target pattern is The discrimination values of the corresponding sections of the reference pattern are compared with each other to determine whether or not they match, thereby recognizing whether or not the embossed characters are correctly imprinted.

However, the shading of the embossed character is affected by various conditions such as the mutual positional relationship between the illuminator, the video camera and the embossed character, the brightness of the illumination light, the influence of disturbance light, and the lack of the embossed character. Is not always constant, so that in the above-mentioned target pattern, there is a section in which the identification value indicating the shading varies for each photographing. As a result, in some cases, the identification value of the target pattern does not match the discrimination value of the reference pattern, despite the fact that the marking is made correctly.

As a countermeasure against this, instead of the above-mentioned reference pattern, a mask pattern which excludes (masks) the section in which the above-mentioned variation occurs from the comparison target is set in advance, and the discrimination value of the mask pattern and the identification value of the above-mentioned target pattern are set. Compare,
It is conceivable to improve pattern recognition by determining whether or not they match.

[0006]

However, when the target pattern greatly fluctuates due to various conditions such as the influence of disturbance light, the discrimination value of the mask pattern can be obtained by comparing the mask pattern with the target pattern. The identification value of the target pattern does not always match.

An object of the present invention is to solve the above problem and to provide a pattern recognition method capable of accurately recognizing a target pattern.

[0008]

In order to achieve the above object, the present invention detects a target pattern by detecting means,
The target pattern is divided, and an identification value corresponding to the gray level of the target pattern is set for each of the obtained regions, and the identification value of each region of the target pattern is divided according to the target pattern. The discrimination value of each area is compared with the discrimination value of each area of the comparison pattern in which the discrimination value is set in advance. In the pattern recognition method for determining a pattern corresponding to the above, as the comparison pattern, a mask pattern in which a preset area in the pattern is excluded from a comparison target between the identification value and the determination value, and all areas in the pattern Sets a plurality of types of reference patterns for comparing the identification value and the determination value, and determines the mask pattern determination value of the target pattern. If the values do not match the different values, the determination value of the reference pattern is compared with the identification value of the target pattern, and the reference value in which the determination value of the reference pattern matches the identification value of the target pattern is the largest. A pattern is selected, and the pattern is determined based on the selected reference pattern.

[0009]

According to the above pattern recognition method, when the judgment value of each area of the mask pattern matches the identification value of the corresponding area of the target pattern, it is determined that the mask pattern is a pattern corresponding to the target pattern. On the other hand, when the determination value of the mask pattern does not match the identification value of the target pattern, the determination value of each region of the plurality of reference patterns is compared with the identification value of the corresponding region of the target pattern, The reference pattern having the largest number of regions where the judgment value of the reference pattern matches the identification value of the target pattern is selected,
By performing pattern discrimination based on the selected reference pattern, the accuracy of pattern recognition is improved, the target pattern can be accurately recognized, and the pattern recognition can be performed quickly.

[0010]

FIG. 1 is a block diagram showing an embodiment of a configuration of a marking station for marking a vehicle number on each vehicle body in an automobile manufacturing line. This engraving station
It is composed of an engraving block 2 for engraving a vehicle number consisting of alphanumeric characters and the like on a vehicle carried into the engraving station, and an engraving recognition block 3 for recognizing the engraved vehicle number. A line controller 10 is connected to the engraving station.
This controls the entire vehicle assembly line.

The marking block 2 includes a marking machine 4.
And a marking controller 5. The marking recognition block 3 includes an illuminator 6, a video camera 7, and an image processing unit 8.
And a display 9 are provided.

The engraving controller 5 comprises a microcomputer or the like. When data such as a vehicle number set according to the type of vehicle is input from the host computer 1 in the factory, the engraving machine 4 is operated based on the data. To control. The engraving machine 4 includes the engraving controller 5
The vehicle number is engraved on a predetermined portion such as a cowl panel in response to a control signal from the vehicle.

The illuminator 6 illuminates the stamped portion of the car body carried into the stamp recognition block 3 with illumination light, thereby shading the stamped character. The video camera 7 (detection means) captures images to detect the engraved characters emitted by the illuminator 6. The image processing unit 8 includes a microcomputer or the like, and recognizes the carved body number imprinted on the basis of a target pattern including the embossed characters captured by the video camera 7. The display 9 displays the recognition result and the like.

Next, the control operation in the marking station will be described with reference to the flowchart of FIG.
First, the production line is started, and the line controller 10 is started.
Is started, the vehicle body is conveyed to the marking block 2. Next, based on data from the host computer 1, a vehicle body number composed of alphanumeric characters and the like is stamped on a predetermined portion of the vehicle body by the stamping machine 4 (step S1).

Thereafter, the vehicle body number (imprinted character) is captured by the video camera 7, and the captured image is taken into the window of the image processing unit 8 for each target pattern (step S2). Subsequently, the image processing unit 8 binarizes each pixel in the window in accordance with shading, and further removes noise (steps S3 and S4).

Next, as shown in FIG. 3, the target pattern T corresponding to the embossed character is divided vertically and horizontally into five sections by the image processing section 8 and divided into 25 sections, and these sections K will be described later. According to the method, it is classified into one of the identification values “0”, “1”, and “2” according to the binarized data in the section, and stored in association with each section K (step S5). S6). The detailed operations of steps S5 and S6 will be described later.

Subsequently, a mask pattern M (see FIG. 4) corresponding to the vehicle number is read from a memory (not shown). This mask pattern M is divided into 25 sections (areas) corresponding to the target pattern T, as shown in FIG. 4, for example, and each of the sections K0 has a predetermined determination value “0”, “1”. , "2" and a determination value "5" indicating a mask section W to be described later.

The image processing section 8 compares the identification value of the section K of the target pattern T with the determination value of the section K0 of the mask pattern M (step S7). In other words, the section K0 of the mask pattern M has the determination values “0”, “1”,
If it is any of “2”, a comparison is made as to whether or not it matches the identification value of the corresponding section K of the target pattern T. When the section K0 is the determination value “5”, the comparison with the section K of the target pattern T corresponding to the mask section W is not performed. The reason why the mask section W is provided in the target pattern T is that the identification value fluctuates for each photographing due to the brightness of illumination light, the influence of disturbance light, and the like. This is because the target pattern T may be erroneously recognized as not corresponding to the vehicle body number even though is normally imprinted. The judgment value of each section K0 of the mask pattern M is set by reading the target pattern of the sample a plurality of times and calculating the total number of binarized data indicating the embossed characters of each section K0. The value “5” indicating the mask section W is set by detecting a section in which the total number of quantified data changes by a predetermined value or more.

Subsequently, it is determined whether or not the determination values of the section K0 other than the mask section W and the identification values of the corresponding section K of the target pattern T all match (step S).
8) When the determination value of each section K0 and the identification value of each section K all match (YES in step S8), the image processing section 8 determines that the embossed character is a correct vehicle number, For example, an OK signal is output to the display 9 (step S9). The display 9 displays that the vehicle number is normally engraved based on the OK signal.

Next, the above-described step S is performed for all the embossed characters.
It is determined whether or not the recognition processing up to 9 has been completed (step S10). If there is any embossed character for which the recognition processing has not been completed (NO in step S10), the process returns to step S5. On the other hand, if the recognition processing up to step S9 has been completed for all the embossed characters, Y is determined in step S10.
It becomes ES, and the vehicle body is transported to another station or the like. Thereafter, the processing from step S1 is performed on the next vehicle body.

On the other hand, the judgment value of the section K0 and the section K
If there is a section where the identification value does not match (NO in step S8), a plurality of reference patterns N corresponding to all the alphanumeric characters and the like that can be engraved as the vehicle body number are sequentially read from a memory (not shown) by the image processing unit 8. It is. For example, as shown in FIG. 5, the reference pattern N is divided into 25 sections corresponding to the target pattern T, and each section K1 has a predetermined determination value “0”, “1”, “1”. 2 "is assigned. Then, the determination value of each section K1 of the reference pattern N is compared with the identification value of each section K of the corresponding target pattern T (step S11). The determination value of each section K1 of each reference pattern N is obtained by reading the target pattern of the sample a plurality of times, obtaining the total number of binarized data indicating the embossed characters in each section K1 for each reading. By averaging the total number of quantified data for each section K1, it is set based on this average value.

Then, among the reference patterns N, a reference pattern N having the largest number of sections where the identification value of the target pattern T matches the determination value of the corresponding reference pattern N is extracted (step S12). Subsequently, it is determined whether or not the most matching reference pattern N is the reference pattern N corresponding to the vehicle number (step S13).
In the case where the above-mentioned most common reference pattern N is the reference pattern N corresponding to the vehicle body number (YE in step S13)
S), the image processing unit 8 determines that the vehicle number is normally imprinted, and proceeds to step S9.

On the other hand, if it is determined in step S13 that the most matching reference pattern N is not the reference pattern N corresponding to the vehicle body number (NO in step S13), whether the processing in steps S3 to S13 has been performed twice or more. It is determined whether or not the above-described execution is the first time (NO in step S14).
Since the discrepancy at step S13 may be caused by the brightness of the illumination light, the influence of disturbance light, or the like, the threshold value for performing the binarization at step S3 is changed (step S15). For example, when it is assumed that the illumination light is darkened due to deterioration of the illumination lamp, the threshold is changed to be small. Thereafter, the process returns to step S3.

On the other hand, if the processing of steps S3 to S13 has been executed for the second time or more in step S14 (YES in step S14), whether the image processing unit 8 has erroneously recognized another embossed character, Since the pattern T cannot be recognized, an NG signal is output to the display 9 (step S16). The display 9 displays that the inscription cannot be recognized according to the NG signal. As a result, the worker recognizes the display and performs processing such as correcting the vehicle body number on the vehicle body.

Next, detailed operations of steps S5 and S6 will be described with reference to the flowcharts of FIGS. That is, the area A of the target pattern T from which noise has been removed by binarization in steps S3 and S4, that is, the total number of binarized data of the target pattern T is calculated (step S21).

Next, whether the area A corresponds to the area in the case of embossed characters, that is, for example, 60 <
It is determined whether or not A <500 (step S2).
2). If the area A does not correspond to the area of the embossed character (NO in step S22), for example, the area A is "60".
In the following cases, noise and the like are mistakenly recognized as embossed characters when the area A is equal to or more than "500", and the process returns to step S3 to process the next target pattern T. Perform

On the other hand, if the area A corresponds to the area of the embossed character (YES in step S22), the target pattern is reduced in order to reduce the influence of shading by illumination light or the like on the recognition accuracy of pattern recognition. A so-called conventionally known thinning process for reducing the line width of the character portion of T is performed (step S23). Subsequently, the area B of the thinned character portion, that is, 2 corresponding to the character portion
The number of the quantified data is calculated (step S24), and it is determined whether or not the area B is equal to or larger than a predetermined value, for example, 15 (step S25). If the area B is less than the predetermined value (NO in step S25), since the target pattern T has erroneously recognized noise or the like as an embossed character, the process returns to step S3 to return to the next target pattern T
Is processed.

On the other hand, if the area B is equal to or larger than the predetermined value (YES in step S25), the area of the target pattern T to be recognized as a character in the target pattern T is further limited in step S26, and the starting coordinates of the image processing are determined ( Well-known feature extraction). Next, in order to divide this area into 25 sections, it is divided vertically and horizontally into five (step S27).

Next, the number of binarized data in each section K is counted (step S28), and the coordinates of each section K are calculated (step S29). Then, the sections K are sequentially taken into the window of the image processing section 8 (step S30), and the area C of the thinned character portion in the section K and the area D of the section K are obtained. The area ratio E (= C / D) is calculated based on the area (steps S31 to S33).

When the area ratio E is equal to or larger than a first threshold value F1 (for example, 10%) set in advance (YES in step S34), an identification value "2" is set in the section K (step S35).

The area ratio E is equal to the first threshold value F.
If it is less than 1 (NO in step S34) and greater than a preset second threshold value F2 (for example, 0%) (YES in step S36), the identification value "1" is set in correspondence with the section K. Is set (step S37). on the other hand,
When the area ratio E is equal to or smaller than the second threshold value F2 (NO in step S36), an identification value “0” is set corresponding to the section K (step S38).

Thereafter, the above step S is performed for all the sections K.
It is determined whether or not the processing from 30 to S38 has been completed (step S39). If there is a section K for which the above processing has not been completed (NO in step S39), step S3 is performed.
Return to 0. On the other hand, the above step S30 is performed for all the sections K.
When the processing of Steps S38 to S38 is completed, YES is determined in Step S39, and the process proceeds to Step S7 described above.

When an identification value is set for each section K of the target pattern T, the process proceeds to step S7, where the target pattern T and the mask pattern M are compared.

As described above, first, the mask pattern M and the target pattern T are compared, and when the judgment value of each section K0 of the mask pattern M matches the identification value of the corresponding section K of the target pattern T, Determines that the embossed character is correctly imprinted, while if the determination value of the section K0 and the identification value of the section K do not match, the determination value of each section K1 of all the reference patterns N And the corresponding target pattern T
Is compared with the identification value of the section K, and the reference pattern N having the largest number of sections in which the determination value of the reference pattern N matches the identification value of the target pattern T is extracted. Since it is determined whether or not the brightness of the illumination light, the influence of disturbance light, etc. can be removed,
Imprinted characters can be accurately recognized by improving the accuracy of pattern recognition, and pattern recognition can be performed more quickly than when only the reference pattern N and the target pattern T are compared. If the reference pattern N with the most matching sections is not the reference pattern N corresponding to the vehicle number, the threshold value for performing the binarization in step S3 is changed and the binarized data of the image captured is changed. Is corrected, the accuracy of pattern recognition is further improved.

In the above embodiment, the mask pattern M is immediately compared with the mask pattern M corresponding to the vehicle body number (imprinted character) in step S7. However, even if the embossed character is an arbitrary alphanumeric character or the like, it can be recognized. it can. In this case, in step S7, the target pattern T and all the mask patterns M corresponding to alphanumeric characters and the like that can be engraved as the vehicle body number are sequentially compared, and the identification value of each section K of the target pattern T and the section of the mask pattern M are compared. When the determination values of K0 match, it is determined that the target pattern T is an alphanumeric character or the like corresponding to the mask pattern M.

Further, in the above embodiment, the case where the pattern recognition of the embossed character is performed is described. However, the present invention is not limited to this, and the case where the pattern recognition of the stamp character, the print character, and the like may be performed.

[0037]

According to the present invention, when the judgment value of the mask pattern does not match the identification value of the target pattern, the reference pattern having the largest number of areas where the judgment value of the reference pattern matches the identification value of the target pattern is determined. Since it is determined that the pattern corresponds to the target pattern, the accuracy of pattern recognition is improved, and the embossed characters can be accurately recognized.
Pattern recognition can be performed quickly.

[Brief description of the drawings]

FIG. 1 is a block diagram showing an embodiment of a configuration of a marking station implementing a pattern recognition method of the present invention.

FIG. 2 is a flowchart showing a control operation in the marking station.

FIG. 3 is an image diagram showing an example of a target pattern.

FIG. 4 is an image diagram showing an example of a mask pattern.

FIG. 5 is an image diagram showing an example of a reference pattern.

FIG. 6 is a flowchart for determining an identification value of a target pattern.

FIG. 7 is a flowchart for determining an identification value of a target pattern.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Host computer 2 Marking block 3 Marking recognition block 4 Marking machine 5 Marking controller 6 Illuminator 7 Video camera (detection means) 8 Image processing unit 9 Display 10 Line controller

──────────────────────────────────────────────────続 き Continued on the front page (58) Field surveyed (Int.Cl. ⁷ , DB name) G06K 9/62 G06T 7/00 JICST file (JOIS)

Claims (1)

(57) [Claims] An object pattern is detected by a detection means,
The target pattern is divided, and an identification value corresponding to the gray level of the target pattern is set for each of the obtained regions, and the identification value of each region of the target pattern is divided according to the target pattern. The discrimination value of each area is compared with the discrimination value of each area of the comparison pattern in which the discrimination value is set in advance. In the pattern recognition method for determining a pattern corresponding to the above, as the comparison pattern, a mask pattern in which a preset area in the pattern is excluded from a comparison target between the identification value and the determination value, and all areas in the pattern Sets a plurality of types of reference patterns for comparing the identification value and the determination value, and determines the mask pattern determination value of the target pattern. If the values do not match the different values, the determination value of the reference pattern is compared with the identification value of the target pattern, and the reference value in which the determination value of the reference pattern matches the identification value of the target pattern is the largest. A pattern recognition method comprising: selecting a pattern; and determining the pattern based on the selected reference pattern.