JP2002312765A - Position recognition method and device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To correctly detect a position even when an object is stained or chipped or the like and to be hardly influenced by chipping and noise. SOLUTION: An image for which the image of the object 1 is picked up is converted so as to extract the contour line 1a, and while retrieving a converted image 3 by a model line 5 set so as to be provided with specified matching point G and non-matching point H in specified position relation with the contour line 1a and to indicate a specified feature amount based on the brightness and by a sub model line 6 set on the inner and outer parts, the feature amount based on the brightness different at the matching point G and the non-matching point H with the contour line 1a is calculated on the model line 5 or the like at each retrieving position and the position of the model line 5 for which the calculated feature amount becomes closest to the specified feature amount is detected as the position of the object 1.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a position detecting method and apparatus for mounting equipment used in a factory, for example, and more specifically, to detecting a position of a low-contrast alignment mark such as an ITO mark of an STN liquid crystal. More particularly, the present invention relates to a position recognition method and apparatus involving image processing suitable for the present invention.

[0002]

2. Description of the Related Art Conventionally, there are various methods for detecting the position of a pattern such as a mark, and the methods are used taking advantage of their respective characteristics. For the detection of the mark position as described above, for example, a template matching method shown in FIG. 16 is often used. This method teaches the position detecting means a template c as shown in FIG. 20 (a) corresponding to a characteristic portion in an image b of the object a shown in FIG. 20 (b), for example, a pattern of a corner portion. deep. The detecting means sequentially searches as shown by arrows in FIG. 20B while comparing each part of the image b with the taught template c. That is, scanning is performed. The position of the target object a is specified and detected at the position of the template c that best matches the portion of the image b as a result of the search. As a method of obtaining the degree of coincidence at the time of retrieval, a normalized correlation method is generally used.

[0003]

However, in the template matching method as described above, if the object a has dirt or chipping, the ratio of those images to noise is high,
It is difficult to accurately detect the position of the object a,
It cannot be detected. To avoid this, there is a countermeasure such as dividing the template c. However, this is still not a sufficient measure due to the influence of the division boundary.

An object of the present invention is to provide a position recognizing method and apparatus capable of correctly detecting a position even if an object has dirt or chipping. It is another object of the present invention to be able to detect a position more rapidly against the influence of chipping and noise and at a high speed. Another object of the invention is to reduce the effects of noise. Another object of the present invention is to enable the position to be correctly detected even if the outline of the object is blurred.

[0005]

In order to achieve the above object, a position recognition method according to the present invention converts an image obtained by capturing an object so as to extract a contour line of the object, and converts the converted image into an image. While searching with a model line that has a specific matching point and a non-matching point in a specific positional relationship with respect to the contour line of the object and indicates a specific feature amount based on their brightness, at each search position On the model line, a feature amount based on different brightness is calculated at a matching point and a mismatching point between the model line and the contour line, and a model in which the feature amount is closest to the specific feature amount in the specific positional relationship is calculated. It is characterized in that a line position is detected as a position of an object.

An apparatus for achieving such a method includes image conversion means for converting an image obtained by capturing an object so as to extract an outline of the object, and an outline of the object on an image converted by the image conversion means. On the model line at each search position, a model line having specific coincidence points and non-coincidence points in a specific positional relationship and indicating a specific feature based on their brightness is searched. Searching means for acquiring different brightness at the coincidence point and the non-coincidence point of the contour line and the outline, and calculating the feature amount based on the acquired brightness, and the feature amount calculated as the search result in the specific positional relationship. And position detecting means for detecting a model line position closest to a specific feature amount as a position of the target object.

[0007] With these arrangements, a contour image of a target object is extracted from a picked-up image of the target object, and the converted image is extracted from the converted image.
A model line that has specific coincidence points and non-coincidence points with the contour of the object and indicates a specific feature based on their brightness is deviated from the specific positional relationship with the contour of the object. Since the feature amount based on the brightness of the coincidence point and the non-coincidence point with the contour line deviates from the specific feature amount, the model line at each search position is searched while the converted image is searched by the model line. And calculating a feature amount based on brightness different between a point of coincidence and a point of non-coincidence with the contour line, and detecting a position of the model line where the calculated feature amount is closest to the specific feature amount as a position of the object. In particular, if the model line coincides or disagrees with the characteristic limited and very small part of the contour of the object, the specific feature amount as described above is obtained. May be dirty or chipped It is hard to appear on specific features and has little effect, and even if there is dirt or noise on the background of the object, the probability of coincidence or disagreement with the model line is very small and this has almost no effect, so detection is accurate. I can do it. In addition, since only the brightness along the model line is acquired and used as the feature amount, even if the brightness of the entire model line is acquired, the handling data is small and the calculation and the detection operation are simple, so the position detection can be performed at high speed. Can be done. In particular, according to the apparatus, stable detection can be automatically performed without error according to a predetermined program.

The position recognizing method according to the present invention further converts an image obtained by capturing the object so as to extract the outline of the object, and specifies the converted image in a specific positional relationship with respect to the outline of the object. While searching with a model line and one or more sub-model lines set inside or outside the model line and having a matching point and a non-matching point, and indicating a specific feature amount based on their brightness, On these model lines and sub-model lines, feature amounts based on different brightness at the coincidence point and the non-coincidence point with the contour line are calculated,
Another feature is to detect, as the position of the object, a model line position at which the characteristic amount is closest to the specific characteristic amount in the specific positional relationship.

An apparatus for achieving such a method includes an image converting means for converting an image obtained by capturing an object so as to extract an outline of the object, and an outline of the object on an image converted by the image converting means. A model line and one or more inside or outside the model line are set to have specific coincidence points and non-coincidence points in a specific positional relationship with the line and to indicate specific feature amounts based on their brightness. While searching with the sub model line, on the model line and sub model line at each search position, different brightness is obtained at the point of coincidence with the contour and at the point of non-coincidence, and the obtained difference in brightness is used as a feature value. A search means for calculating, and a model line position at which the characteristic amount calculated as a result of the search is closest to the characteristic amount due to the specific brightness difference in the specific positional relationship is set as the position of the object. Position detecting means for output, it Sonaere a.

In these configurations, specific features based on differences in brightness at points of coincidence and non-coincidence with the contour line of the object are increased by the number of sub-model lines set inside or outside the model line. Since the amount of individuality increases, the position of the contour including the discrimination of the contour of a different object is more easily detected, and the position is detected more accurately without a decrease in the detection speed even though the number of data is slightly increased. It is possible to cope with the case where the outlines of a plurality of different objects are mixed. In particular, according to the apparatus, the detection of the error-free position can be stably and automatically performed according to a predetermined program.

In this case, the feature quantity is the ratio of the number of brightness positions corresponding to the contour line on the model line to the number of brightness positions corresponding to the contour line on the sub model line. But the brightness on the model line, the brightness on the sub model line,
The ratio of numbers has the advantage that the number of data is small, and the ratio of brightness allows further individualization of a specific feature, so that the accuracy of detection is further improved. .

[0012] Further, blurring of the object or its outline on the image before and after the conversion is detected, and if there is a blurring, the sub model line is moved out of the range where the blurring is generated or newly set and set. By using this, the position can be detected using the model line and the sub model line without the influence of blur.

In order to achieve such a method, the apparatus includes a blur detecting means for detecting blur of an object or its outline in an image before and after the conversion, and an area in which the blur detected by the blur detecting means has occurred. Model line changing means for changing the sub model line so as to move or newly set the sub model line and use it may be provided.

[0014] Further, since the image conversion is to convert the brightness on the model line and the sub model line after performing the edge emphasizing operation, the object or the background may have chipping, dirt or noise. This can be easily excluded from the image and its brightness data, and the position can be detected accurately and at high speed.

In order to achieve such a method, the apparatus includes, as image conversion means, edge enhancement means for performing edge enhancement operation, and density conversion means for density conversion of brightness on the model line and the sub model line. I just need.

Further features and operations of the present invention will become apparent from the following detailed description and drawings. Each feature of the present invention is solely as much as possible,
Alternatively, various combinations can be used in combination.

[0017]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, with reference to the accompanying drawings, a position recognition method and apparatus according to some embodiments of the present invention will be described in detail to provide an understanding of the present invention. The following embodiments are specific examples of the present invention and do not limit the technical scope of the present invention.

(Embodiment 1) As shown in FIG. 1, the position recognition method according to the first embodiment first uses an object 1 as shown in FIGS. A converted image 3 obtained by extracting the two outlines 1a and 2a is obtained. Next, the outline 1 of the objects 1 and 2 is displayed on the converted image 3.
2a, a specific matching point G and a non-matching point H in a specific positional relationship such that the centers match each other as exemplified in FIGS. 2A and 2B, and identification based on their brightness. While sequentially searching using the model line 5 set so as to indicate the feature amount of the model line 5, the model line 5 at each search position is searched.
Above, the feature amount based on the brightness different at the coincidence point G and the non-coincidence point H with the contour lines 1a and 2a is calculated, and the position of the model line 5 where the feature amount is closest to the specific feature amount is calculated. It is detected as the position of the object 1 or 2.

FIG. 2 shows the operation of extracting the contour lines 1a and 2a of the objects 1 and 2 from the captured images of the objects 1 and 2 as described above.
(A) On the converted image 3 as shown in (b), specific features corresponding to the specific coincidence points G and non-coincidence points H with respect to the contour lines 1a and 2a of the objects 1 and 2 and based on their brightness. For example, the model line 5 set so as to indicate the position of the specific brightness indicated by the coincidence point G or the non-coincidence point H, the number thereof, or their ratio, etc.
a with respect to a specific positional relationship as shown in FIGS.
Here, as the center of the model line 5 and the center of the contour lines 1a and 2a deviate from the coincident positional relationship, the feature based on the brightness of the coincidence point G and the mismatch point H between the model line 5 and the contour lines 1a and 2a. The amount deviates from the specific feature amount.
Therefore, while searching the converted image 3 using the model line 5, the model line 5 and the contour lines 1a, 2a at each search position are obtained.
Is calculated based on the brightness different between the coincidence point G and the non-coincidence point H, and the position of the model line 5 where the calculated characteristic amount is closest to the specific characteristic amount is determined. It can be detected as a position.

Here, as shown in FIGS. 2 (a) and 2 (b), the model line 5 coincides or disagrees with the contour lines 1a and 2a of the objects 1 and 2 in a limited and characteristically small portion. Therefore, even if the objects 1 and 2 are dirty or chipped, they are difficult to appear in the specific features and hardly affect the objects 1 and 2. Even if there is dirt or noise in the background, the probability of coincidence or non-coincidence with the model line is extremely small and has little effect, so that detection can be performed accurately. Further, since it is only necessary to obtain the brightness along the model line 5 and use it as a feature amount, the model line 5
Even if the brightness of the entire region is acquired, the handling data is small and the calculation and the detection operation are simplified, so that the position detection can be performed at high speed. Note that the thickness of the model line 5 may be such that pixels are arranged one by one, but it is preferable to match the pixel width of the contour lines 1a and 2a because it is easy to determine coincidence or non-coincidence. However, it is not limited to this.

The model line 5 in the first embodiment is set to have the same size and shape as the circular contour 2a of the object 2 shown in FIG. Unlike the case of the template, the coincidence point G when both concentrically overlap each other is halved, and there is no mismatch point H. The model line 5 is also the one in which the outline 1a of the object 1 shown in FIG. 2A is circumscribed, and the coincidence point G when both are concentrically overlapped is further than that of the circular outline 2a. Less,
There is a non-coincidence point H, but even less than the coincidence point G. Then, as the model line 5 is displaced from the contour lines 1a and 2a, the relationship between the matching point G and the mismatching point H between the model line 5 and the contour lines 1a and 2a changes. There is almost no probability of coincidence between lines 1a and 2a.

Therefore, a specific feature amount in a specific positional relationship concentrically overlapping the contour line 1a of the model line 5 is determined by using the mismatch point H
And a specific feature amount in a specific positional relationship concentric with and overlapping the contour line 2a is represented by a mismatch point H.
Even if the difference between the contour lines 1a and 2a is large and the influence of noise is large, the feature amount obtained at each search position is the most. The contour line 1a,
The position of 2a, that is, the positions of the objects 1, 2 can be correctly detected. Thereby, even if the contour lines 1a and 2a of different objects 1 and 2 are mixed in the same converted image 3,
Both positions can be detected while determining the difference between the objects 1 and 2. However, these contours 1a, 2a
Are on the different transformed image 3, they can of course be individually detected by the model line 5.

If these objects 1 and 2 are position recognition marks such as a liquid crystal substrate whose position is to be recognized, the position of the liquid crystal substrate or the like whose position is to be recognized is determined based on the positions detected for the objects 1 and 2. Can be detected. Note that the model line 5 does not need to be in a regular discontinuous state, and is not limited to a discontinuous line and may be a continuous line.

As such a position recognizing device, as shown in FIG. 3, an image conversion means 31 for converting an image of the object 1 or 2 so as to extract the outlines 1a and 2a of the object 1 or 2 as shown in FIG. And the converted image 3 by the image converting means 31
Is sequentially searched with the model line 5 set so as to show a specific degree of coincidence and a degree of non-coincidence in the specific positional relationship as described above with respect to the contour lines 1a, 2a, etc. of the objects 1, 2, and so on. A search unit 32 that acquires a specific brightness that differs between a match point G and a mismatch point H between the model line 5 and the contour lines 1a and 2a in the state, and calculates a feature amount based on the obtained brightness; Position detecting means 36 for detecting the position of the model line 5 where the calculated characteristic amount is closest to the specific characteristic amount as the position of the objects 1 and 2 may be provided. The search means 32 includes a brightness obtaining means 33 for obtaining the brightness on the model line 5 at each search position, and a calculation means 35 for calculating a feature amount based on the brightness obtained at each search position. Be composed. These means can be an internal function of a computer or a dedicated computer for controlling the operation of the liquid crystal substrate manufacturing equipment which needs the position detection as described above, and a part or all of the means can be provided as an individual device dedicated to them. can do.

When the contour line 2a of the circular object 2 appears only discontinuously as shown in FIG. 4B, when the model line 5 concentrically overlaps with the model line 5, the coincidence points G and The number of non-coincidence points H is the same as the relationship between the model line 5 and the rectangular outline 1a shown in FIG. 4A, and the feature amounts calculated on both sides coincide. This means that when it is desired to individually detect the positions of the outlines 1a and 2a on the same converted image 3, the two cannot be distinguished. Although the position information of the mismatching point H can be distinguished, the operation becomes complicated and the inspection speed becomes slow. Embodiment 2 below can deal with such a problem.

(Embodiment 2) In this embodiment 2, as shown in FIG. 5, first, objects 1 and 2 as shown in FIGS.
Are contour lines 1a and 2a of the objects 1 and 2
The converted image 3 converted so as to extract
The model line 5 and the inside of the model line 5 are set so as to have a specific matching point G and a non-matching point H in a specific positional relationship with respect to the contour lines 1a and 2a, and indicate a specific feature amount based on their brightness. Or while sequentially searching for one or more sub model lines 6 set on the outside, and matching points G between these model lines 5 and sub model lines 6... And contour lines 1a, 2a at each search position. While acquiring different brightnesses at the mismatch point H, the difference in the acquired brightness is calculated as a feature amount. Next, the position of the model line 5 where the characteristic amount is closest to the specific characteristic amount is detected as the position of the objects 1 and 2.

In this manner, as compared with the case where only the model line 5 of the first embodiment is used, the number of the sub-model lines 6... Set inside or outside the model line 5 increases. , The individuality of a specific feature value due to the difference in brightness at the point of coincidence G and the point of non-coincidence H with the outlines 1a, 2a is increased. For example, in FIG. 6A and FIG. 6B, the coincidence point G and the non-coincidence point H with respect to the contour lines 1a and 2a detected by the model line 5 are the same in number and cannot be distinguished from each other. According to the line 6, there are six coincidence points G with respect to the contour line 1a in FIG. 6A, and there is no coincidence point G with the contour line 2a. And differentiated position detection can be performed. Specifically, when the model line 5 is viewed from the number of brightness at the mismatch point H and the sub model line 6 is viewed from the number of brightness at the match point G, it is easy to differentiate with a small number of data.

As an apparatus for achieving such a position recognition method, as shown in FIG. 7, an image obtained by capturing the objects 1 and 2 is converted so as to extract the contour lines 1a and 2a of the objects 1 and 2. The image conversion means 31 and the converted image 3 by the image conversion means 31 have a specific coincidence point G and a non-coincidence point H in the specific positional relationship as described above with respect to the contour lines 1a, 2a, etc. of the objects 1, 2. The model line 5 and the sub model line 6... Set so as to indicate a specific feature amount based on the brightness are sequentially searched, and at each search position, the model line 5 and the sub model line 6. Retrieval means 3 for acquiring different brightnesses at a point of coincidence G and a point of non-coincidence H with lines 1a and 2a and calculating a difference in the acquired brightness as a feature amount.
2 and a position detecting means 36 for detecting the position of the model line 5 where the calculated characteristic amount is closest to the specific characteristic amount as the positions of the objects 1 and 2.
2 is a brightness acquisition means 3 for acquiring the brightness on the model line 5
In addition to 3, a brightness acquisition unit 34 for acquiring the brightness on the sub model line 6 is provided, and the feature amount is calculated by the calculation unit 35 based on the brightness information from the brightness acquisition units 33 and 34. These means can be an internal function of a computer or a dedicated computer for controlling the operation of the liquid crystal substrate manufacturing equipment which needs the position detection as described above, and a part or all of the means can be provided as an individual device dedicated to them. can do.

On the other hand, as shown in FIG.
Considering the case where the number of mismatch points H between the model line 5 and the discontinuous contour line 2a of the object 2 is greater than the case shown in FIG. The tendency that the degree of inconsistency of the line 5 with respect to the outlines 1a and 2a is greater than the degree of coincidence of the sub model line 6 with the outlines 1a and 2a is similar. However, the image in which the sub model line 6 coincides around the contour line 2a in FIG. 8B is a background stain or other noise 11, the number of existing sub models and the probability of coincidence with the sub model line 6. Is much smaller than that for the contour line 1a of the sub model line 6 in FIG. Such brightness acquisition states are shown in Tables 1 and 2 below.

[0030]

[Table 1]

[0031]

[Table 2] Therefore, if such a tendency is regarded as a feature amount, it is possible to perform position detection that distinguishes both the outlines 1a and 2a in FIGS. 8A and 8B. The third embodiment takes such measures.

(Embodiment 3) This embodiment 3 treats the ratio of the brightness on the model line 5 to the brightness on the sub model line 6 as a feature quantity, so that the model line 5, sub model line 6,. The specific feature value is further individualized when the concentricity coincides with the line 1a and when the concentricity coincides with the contour line 2a. Specifically, as shown in FIG. 9, first, contour lines 1a and 2a of the objects 1 and 2 are extracted from images obtained by capturing the objects 1 and 2 as shown in FIGS. Image line 5 set so as to show specific coincidence points G and non-coincidence points H in a specific positional relationship with respect to contour lines 1a and 2a of objects 1 and 2
And the secondary model line 6 set inside or outside the model line 5 to sequentially search, and at each search position, the coincidence point G between the model line 5 and the secondary model line 6... And the contour lines 1a, 2a. And different brightness at the mismatch point H
The number of brightness positions that are coincident with the contour lines 1a and 2a of the sub model line 6 is defined as the denominator, and the contour lines 1a and 2
The ratio of the number of brightness positions that are the coincidence point G with the numerator is calculated as a feature amount, and the calculated feature amount is the model line 5, the sub model line 6, the model line 5 and the contour lines 1a, 2a.
The position of the model line 5 that is closest to a specific feature value obtained by taking the ratio of the number of brightness positions that become the coincidence point G with the numerator as the position of the object 1 or 2 is detected.

For example, in the example of FIG. 8A, the specific feature amount is 13/6 from the feature amount data shown in Table 1, and FIG.
In the example of (b), since the specific feature amount is 8/1 from the feature amount data shown in Table 2, it is possible to obtain a specific feature amount in which the differentiation between the two is further increased. Therefore, the influence of the noise 11 such as dirt on the background of the object 2 is small and the difference between the outlines 1a and 2a is clearly determined while the outline 1
The positions a and 2a can be accurately detected.

This is the same as the background noise 11 with respect to noise due to dirt or chipping of the object 2 inside the contour line 2a.
This can be coped with by providing the sub model line 6 inside the model line 5. Further, if the sub model line 6 is provided inside and outside the model line 5, it is possible to cope with the noise on the object 2 and the background noise. However, a set of the model line 5 and the outside sub model line 6 and the model line 5 and inner submodel line 6
It is preferable to individually calculate and compare the feature amounts with the set of in order to minimize the influence of noise.

As an apparatus for achieving such a position recognition method, as shown in FIG. 10, conversion is performed so that contour lines 1a and 2a of the objects 1 and 2 are extracted from an image obtained by capturing the objects 1 and 2. Image conversion means 31 and this image conversion means 3
1 is converted to the contour lines 1a, 2a of the objects 1, 2
The model line 5 and the model line 5 are set so as to have a specific matching point G and a non-matching point H in a specific positional relationship so as to indicate a specific feature amount based on their brightness. Or a plurality of sub model lines 6... Are sequentially searched, and the coincidence point G between the model line 5 at each search position and the contour lines 1 a and 2 a on the sub model line 6.
Searching means for separately acquiring the brightness different between the model line 5 and the sub model line 6 at the non-coincidence point H and calculating the specific inside / outside ratio in the characteristic amount based on the acquired brightness as the specific characteristic amount 32, and a position detecting unit 36 that detects the position of the model line 5 where the calculated characteristic amount is closest to the specific characteristic amount as the position of the object 1 or 2. Brightness acquiring means 33 and 34 for acquiring the brightness on the model line 5 and the sub model line 6, and an inside / outside ratio for calculating a specific inside / outside ratio as a feature amount from the brightness information from the brightness acquiring sections 33 and 34. What is necessary is just to comprise with the calculation means 37.

(Embodiment 4) In Embodiment 4, as shown in FIG. 11, first, an object 1 as shown in FIGS.
2 are contour lines 1a and 2 of the objects 1 and 2
After converting the image with the edge enhancement operation to extract a,
The brightness on the model line 5 and the sub model line 6 in the converted image obtained thereby, that is, their contours 1a,
The density conversion is performed on the brightness at the coincidence point G with 2a and the other non-coincidence point H. Next, the converted image 3 converted in this way is provided with a specific feature G based on their brightness, having a specific matching point G and a non-matching point H in a specific positional relationship with respect to the contours 1a, 2a of the objects 1, 2. Are sequentially searched for the model line 5 and the sub-model lines 6 set inside or outside the model line 5 or one or a plurality of sub-model lines 6... While obtaining different brightnesses at the coincidence point G and the non-coincidence point H between the contours 1a and 2a, the feature amount based on the acquired difference in brightness is calculated. Subsequently, the position of the model line 5 where this feature amount is closest to the specific feature amount is set to the object 1, 2
Is detected as the position.

By performing the edge emphasizing operation and the density conversion as described above, even if the objects 1, 2 and the background have chipping, dirt, or noise, they can be easily excluded from the image or its brightness data, and the position can be more accurately determined. Can be detected.

The edge enhancement operation is performed by using a Laplacian filter 21 as shown in FIG. 12 or by using SOBEL filters 22 and 23 for XY enhancement as shown in FIGS. Laplacian filter 2 shown in FIG.
In the case of 1, the central value is applied to the pixel of interest to increase the density by + α.
Then, the operation of setting the density of the peripheral pixels to -β is performed for the entire screen and the absolute value is output, so that a portion having a strong edge is extracted brightly. In the illustrated example, α = 8 and β = −1. In the case of the XY-enhancing SOBEL filters 22 and 23 shown in FIGS. 13 and 14, addition and subtraction for emphasizing the edge in the X direction by the X-enhancing SOBEL filter 22 are performed by numerical settings as illustrated in FIG. 13. The pixels in the target direction and the pixels in the target direction are added to and subtracted from the pixels in the target direction and its peripheral pixels by numerical setting as shown in FIG. By performing with the neighboring pixels and outputting the magnitude of the result of the addition and subtraction, a portion having a strong edge is extracted brightly.

Next, in the converted image 3 as shown in FIG. 8, the ratio of the total brightness of the contour lines 1a and 2a on the model line 5 and the sub model line 6 is replaced with the ratio between the inside and outside ratios in the third embodiment. Consider a case where a specific feature amount is used. Normally, the objects 1 and 2 and their outlines 1a and 2a are extracted brightly, and if the brightness is 30 and the brightness of the other background portion is 5, the brightness of the coincidence point G is 30, and the non-coincidence point is The brightness of H is 5. This is summarized for each brightness acquisition point as shown in Tables 3 and 4 below. In Table 4, the case where the set brightness has changed due to image deterioration or the like is shown in parentheses.

[0040]

[Table 3]

[0041]

[Table 4] Here, the ratio between the total brightness at the coincidence point G and the non-coincidence point H on the model line 5 and the total brightness at the coincidence point G and the non-coincidence point H on the sub model line 6 in FIG. I do.

According to this, the total brightness of the detection points on the model line 5 for the contour line 1a shown in Table 3 is ((5 × 5)
+ (13 × 30)) = 415, and the total brightness of the detection points on the sub model line 6 is ((18 × 5) + (6 × 30)) = 270
It is. Therefore, these ratios are 415/270 = 1.5
4 (Equation 1), which is a specific feature amount for detecting the position of the contour line 1a from the model line 5 and the sub model line 6 while determining the difference from the contour line 2a.

The total brightness of the detection points on the model line 5 for the contour 2a shown in Table 4 is ((10 × 5) +
(8 × 30)) = 290, and the total brightness of the detection points on the sub model line 6 is ((23 × 5) + (1 × 30)) = 145. Therefore, these ratios are 290/145 = 2 (Equation 2)
The model line 5 and the sub model line 6 serve as a specific feature amount for detecting the position of the contour 2a while determining the difference from the contour 1a.

Thus, the model line 5 and the sub model line 6
Even if the above total brightness ratio is set to a specific feature amount, the positions of the different contours 1a, 2a of the different objects 1, 2 using the same model line 5 and the sub model line 6 are determined. For detection, the overlapping state of the same model line 5 and sub model line 6 with each contour line 1a, 2a is different, and the point of coincidence between the same model line 5 and sub model line 6 and each contour line 1a, 2a is different. The difference in the number and distribution of G and mismatch points H results in the total brightness of the detection points on the model line 5 and the sub model line 6
The ratio between the total brightness of the upper detection points and the contour lines 1a and 2
a, the contour line 1a
And 2a, and the positions of the model line 5 and the sub model line 6 and the contour lines 1a and 2a can be detected based on the coincidence conditions of the positions.

However, the converted image in FIG. 8B deteriorates, and the brightness of the contour line 2a decreases from 30 to 25 and the brightness of the noise 11 decreases from 30 to 50 as shown in the parentheses in Table 4. When the brightness ratio is recalculated for the case of
The specific feature is ((10 × 5) + (8 × 25)) /
((23 × 5) + (1 × 50)) = 1.52 (Equation 3), and the specific feature amount 1.1 for the contour line 1a shown in Table 3.
54 slightly smaller than the contour lines 1a, 2
a cannot be distinguished.

Therefore, the model line 5 and the sub model line 6
Such a problem can be solved by performing the density conversion on the upper brightness, that is, each of the coincidence point G with the contour lines 1a and 2a and the other non-coincidence point H. The simplest density conversion is a threshold operation for binarization. For example, assuming that the threshold value 15 is binarized and the brightness of 15 or more is 1 and the brightness of less than 15 is 0, the inside / outside ratio of the brightness in the detection state of FIG. 8A is ((5 × 0) + (13 × 1)) / ((6 ×
1) + (18 × 0)) = 13/6 = 2.2 (Equation 4).

On the other hand, irrespective of the brightness in (Expression 2) and (Expression 3) in the detection state of FIG.
The ratio of brightness to exterior is ((10 × 0) + (8 × 1)) /
((1 × 1) + (23 × 0)) = 8/1 = 8 (Equation 5).

By such a thresholding operation for binarization, a feature amount based on brightness that can accurately determine the contour lines 1a and 2a can be obtained. Note that the binarized threshold here has an absolute brightness component that has already been eliminated by edge emphasis, so that an effect can be obtained even with a fixed threshold. If it is necessary to enhance the effect, a threshold value may be automatically obtained from a histogram relating to brightness or the like. Furthermore, instead of the above-described binarization, it is also possible to cope with an operation such as keeping the brightness above a certain threshold constant and leaving the density component otherwise.

As described above, in the fourth embodiment, the image is converted so as to extract the contour lines 1a and 2a, and the brightness on the model line 5 on the converted image 3 and the 1 or 2 on the inside or outside of the model line 5 are converted. The feature amount calculated by performing the sequential conversion image search while measuring the brightness on the plurality of set sub model lines 6 is closest to the specific feature amount in relation to the contour lines 1a and 2a of the objects 1 and 2. Is used as a detection position, in addition to enabling accurate and high-speed position detection even when there is dirt on the objects 1 and 2 and dirt on a chipped background, the contour lines 1a and 2a are subjected to edge enhancement operation. In addition to the extracted ones, the density conversion is performed on the brightness distribution detected on the model line 5 and the sub model line 6 of the contour lines 1a and 2a, so that the position can be detected more quickly due to chipping or noise. With , It is possible to further enhance the less to the detection accuracy of the influence of noise and edge intensity fluctuations.

An apparatus for achieving such a method is shown in FIG.
The image conversion unit 31 of the apparatus shown in FIG. 10 may include an edge enhancement unit 38 for performing an edge enhancement operation, and a density conversion unit 39 for converting the brightness of the contour lines 1a and 2a and other brightness. Since there is no difference from the configuration of the apparatus shown in FIGS. 7 and 10, duplicate description will be omitted. This can also be applied to the devices shown in FIGS.

(Fifth Embodiment) In the fifth embodiment, when the contour images of the objects 1 and 2 are blurred, this is determined to enable accurate position detection. Detection of blur may be performed before or after conversion of the captured image. Before the conversion, the data may be obtained from, for example, data taught in advance, or may be automatically determined from the converted image 3 as shown in FIGS. FIG.
The contour 2a in the converted image 3 of the object 2 as shown in FIG. 6A has no blur, and the position detection described with reference to FIGS. 6 and 8 can be performed. However, in the converted image 3 shown in FIG. 16B, the outline 2a of the object 2 is blurred. Contour 2
When blur occurs in a, the outline 2a becomes thicker inward or outward than in the case of FIG. 16A, as shown in FIG. 16B. In the example of FIG. 16B, blur is generated on the outside.
As a result, the discontinuity of the contour line 2a as shown in FIG. 16A is larger than the coincidence of the sub-model line 6 with the model line 5, or conversely, the discontinuity of the contour line 2a is not coincident with the model line 5. Since the degree of inconsistency of the sub-model line 6 with respect to the degree increases, or the degree of coincidence with the sub-model line 6 becomes much greater than the degree of inconsistency with respect to the model line 5, a characteristic change occurs. The presence or absence of blurring of the outline 2a can be determined from the converted image 3 based on the presence or absence of.

If there is a change in the characteristic amount due to the blur, the contour line 1a is formed by the model line 5 shown in the second to fourth embodiments and one or more sub model lines 6 inside or outside the model line 5. It becomes difficult or impossible to detect those positions while discriminating between 2a and 2a. However, as shown in FIG. 16C, by moving the sub model line 6 out of the range of the blur of the contour line 2a or by newly setting it, the sub model line 6 is used as in each of the embodiments 2 to 4. The position detection accompanying the distinction of the characteristic contour lines 1a, 2a can be performed.

Therefore, in the fifth embodiment, as shown in FIG. 17, for example, the presence or absence of blur is detected before image conversion. Next, the image is converted so as to extract the outline 2a, but if blurring has occurred, the sub model line 6 is moved or newly set so as to be located outside the blurring range as described above. More specifically, by moving the sub model line 6 to the outside of the range of the blur occurring outside the contour line 2a or newly setting it and using it, the model line 6 is not affected by the blur. 5 and the position detection using the sub model line 6 can be performed in the same manner as in any one of the second to fourth embodiments.

FIGS. 18A and 18B show another two methods for detecting blur. The method shown in FIG.
From the relationship between each pixel position traversing the contour 2a and the density at each pixel position, the gradient of the density between the flat portion corresponding to the object 2 and the flat portion corresponding to the background is the blur of the contour 2a. , And detects a pixel range in which the gradient of density extends, as a blur amount. In the method shown in FIG. 18B, the slope or curve until the density at each pixel position crossing the contour portion 2a coincides with the pixel density on the model line 5 is changed to the contour line 2.
The pixel range corresponding to the blur of a and covered by the slope or curve is detected as the blur amount.

As described above, the detection of the blur may be performed at the time of teaching or at the time of recognizing the position detection. In particular, when the blur is obtained by the teaching, if the blur is too large, it can be treated as an error.

In the fifth embodiment, the image is converted so as to extract the contour line 2a, and the brightness of the contour line 2a of the object 2 on the model line 5 on the converted image 3 and the brightness inside or on the model line 5 are determined. A feature amount is calculated from the brightness on one or more sub-model lines 6 set outside, and the position of the model line 5 where the calculated feature amount is closest to the specific feature amount is detected as the position of the object. In addition to enabling accurate and high-speed position detection even when the objects 1 and 2 are stained or a missing background is stained, the blur of the object 2 or the outline 2a of the image before and after the conversion is detected. By moving or newly setting the sub model line 6 outside the range of the detected blur, and using it in combination with the model line 5, even if the outline of the object or the outline 2a is blurred, it is not affected. To correct position detection or different Possible to distinguish between such contour 1a.
Further, it is also possible to determine good or bad according to the blur amount.

As shown in FIG. 19, a device for achieving such a method is a blur detecting means 40 for detecting blur of the object 2 in an image of the object 2 and a blur detecting means 40 for detecting the blur. The sub model line 6 may be provided outside the range in which the blur has occurred, or the sub model line changing means 41 may be provided to change the sub model line 6 so that the sub model line 6 is set and newly used. 7 and the configuration of the apparatus shown in FIG.

[0058]

According to the position recognition method and apparatus of the present invention, on the converted image obtained by performing the operation of extracting the outline of the object from the captured image of the object, a specific coincidence point and a specific point with respect to the outline of the object are obtained. A model line that has a mismatch point and is set to indicate a specific feature amount based on their brightness, as the position of the model line deviates from a specific positional relationship with the contour line of the object, the point of coincidence with the contour line and the mismatch point. Since the feature amount based on the brightness of the point deviates from the specific feature amount, while searching for the converted image using the model line, the matching point and the mismatching point between the model line and the contour line at each search position are determined. By calculating a feature amount based on different brightness, the position of the model line where the calculated feature amount is closest to the specific feature amount can be detected as the position of the target object. Features with contour lines The specific feature as described above can be obtained if they match or disagree with a limited and extremely small part, so even if the object has dirt or chipping, it is difficult to appear in the specific feature. It has little effect, and even if there is dirt or noise on the background of the object, the probability of coincidence or non-coincidence with the model line is extremely small, and this has little effect, so that accurate detection can be performed. In addition, since only the brightness along the model line is acquired and used as the feature amount, even if the brightness of the entire model line is acquired, the handling data is small and the calculation and the detection operation are simple, so the position detection can be performed at high speed. Can be done. In particular, according to the apparatus, stable detection can be automatically performed without error according to a predetermined program.

According to another position recognition method and apparatus of the present invention, the coincidence points and the non-coincidence points with the contour of the object are increased by the number of sub-model lines set inside or outside the model line. Since the individualization of specific features due to differences in brightness increases, the position of the outline including the determination of the outline of different objects becomes easier to detect, and the number of data increases somewhat, but the detection speed decreases. It is possible to detect the position more accurately and to cope with the case where the outlines of a plurality of different objects are mixed. In particular, according to the apparatus, the detection of the error-free position can be stably and automatically performed according to a predetermined program.

In this case, the feature quantity is the ratio of the number of brightness positions where the feature point coincides with the contour line on the model line to the number of brightness positions where the feature point coincides with the contour line on the sub model line. But the brightness on the model line, the brightness on the sub model line,
The ratio of numbers has the advantage that the number of data is small, and the ratio of brightness allows further individualization of a specific feature, so that the accuracy of detection is further improved. .

Also, the blur of the object and its outline on the image before and after the conversion is detected, and if there is a blur, the sub model line is moved out of the range where the blur is generated or newly set and set. By using this, the position can be detected using the model line and the sub model line without the influence of blur.

Further, since the image conversion is to convert the brightness on the model line and the sub model line after performing the edge emphasizing operation, the object or the background may be missing, dirt, or noise. This can be easily excluded from the image and its brightness data, and the position can be detected accurately and at high speed.

In order to achieve such a method, the apparatus includes, as image conversion means, edge enhancement means for performing edge enhancement operation, and density conversion means for density conversion of brightness on the model line and the sub model line. I just need.

Further features and operations of the present invention will become apparent from the following detailed description and drawings. Each feature of the present invention is solely as much as possible,
Alternatively, various combinations can be used in combination.

Further features and operations of the present invention will become apparent from the following detailed description and drawings. Each feature of the present invention is solely as much as possible,
Alternatively, various combinations can be used in combination.

[Brief description of the drawings]

FIG. 1 is a flowchart illustrating a position recognition method according to a first embodiment of the present invention.

FIG. 2 is an explanatory diagram schematically showing a converted image in which a contour of an object is extracted in a case where the method of FIG. 1 is applied, and a model line for detecting a feature amount for searching a position of the contour; .

FIG. 3 is a block diagram of a position recognition device that performs the method of FIG. 1;

FIG. 4 is an explanatory diagram showing a case in which a different contour cannot be distinguished by a method using only model lines in FIG. 1;

FIG. 5 is a flowchart illustrating a position recognition method according to a second embodiment of the present invention.

FIG. 6 schematically shows a converted image obtained by extracting the contour of the object when the method of FIG. 5 is applied, and a model line and a sub-model line for detecting a feature amount for searching for the position of the contour. FIG.

FIG. 7 is a block diagram of a position recognition device that achieves the method of FIG. 5;

FIG. 8 shows that the method of FIG. 5 makes it impossible to discriminate different outlines, and obtains the ratio of the number of coincidence points between the model line and the sub model line to the outline for each different outline as a solution to the problem. It is explanatory drawing which shows typically the position recognition method to compare.

FIG. 9 shows another solution to the case where it becomes impossible to discriminate a different contour line by the method of FIG. 5, and obtains the ratio of the total brightness on the model line to the total brightness on the sub model line for each different contour line. 11 is a flowchart illustrating a position recognition method according to a third embodiment of the present invention, in which position comparison is performed and position detection is performed.

FIG. 10 is a block diagram of a position recognition device that achieves the method of FIG. 9;

FIG. 11 is a flowchart illustrating a position recognition method according to a fourth embodiment for performing edge enhancement and density conversion according to the present invention.

FIG. 12 is an explanatory diagram schematically illustrating an example of a Laplacian filter that performs an edge enhancement operation.

FIG. 13 is an explanatory diagram illustrating an example of a SOBEL filter for X enhancement that performs an edge enhancement operation.

FIG. 14 is an explanatory diagram illustrating an example of a SOBEL filter for Y enhancement that performs an edge enhancement operation.

FIG. 15 is a block diagram of a position recognition device that achieves the method shown in FIG.

FIG. 16 is a diagram showing a situation in which it becomes difficult or impossible to determine a different contour line or detect a position when a contour or a contour line of an object is blurred, and a position of a sub model line as a solution to the problem. (A) is a state without blur, (b) is a state where blur occurs, and (c) is a sub model according to the blur. The line has been changed.

FIG. 17 is a flowchart illustrating a position recognition method according to a fifth embodiment of the present invention to which the solution shown in FIG. 16 is applied.

FIG. 18 is a graph showing another two blur detection methods.

FIG. 19 is a block diagram of an apparatus for performing the method of FIG. 18;

FIG. 20 is an explanatory diagram showing a conventional template-based position recognition method.

[Explanation of symbols]

 1, 2 Object 1a, 2a Contour 3 Converted image 5 Model line 6 Submodel line 21 Laplacian filter 22, 23 SOBEL filter 31 Image conversion means 32 Search means 33 Brightness acquisition means on model line 34 Brightness on submodel line Acquisition means 35 Calculation means 36 Position detection means 37 Interior / outside ratio calculation means 38 Edge enhancement means 39 Density conversion means 40 Blur detection means 41 Sub model line changing means G Match point H Mismatch point

 ──────────────────────────────────────────────────続 き Continued on the front page (72) Inventor Takayuki Fukae 1006 Kazuma Kadoma, Osaka Prefecture Matsushita Electric Industrial Co., Ltd. Terms (reference) 2F065 AA03 AA14 AA56 BB02 BB28 CC25 DD06 DD13 FF04 FF26 QQ00 QQ04 QQ24 QQ26 QQ27 QQ33 QQ39 UU05 5B057 AA03 CA08 CA12 CA16 DA07 DB02 DB09 DC16 DC22 DC36 5L096 AA69 GA04 FA06

Claims (14)

[Claims] 1. An image obtained by capturing an image of an object is converted so as to extract an outline of the object, and the converted image has specific coincidence points and non-coincidence points in a specific positional relationship with respect to the outline of the object. While searching with a model line set to indicate a specific feature based on those brightnesses, the brightness at the matching and non-matching points between the model line and the contour line on the model line at each search position is different And calculating a feature amount based on the model line, and detecting a model line position at which the feature amount is closest to the specific feature amount in the specific positional relationship as an object position. 2. An image obtained by capturing an image of an object is converted so as to extract an outline of the object, and the converted image has specific coincidence points and non-coincidence points in a specific positional relationship with respect to the outline of the object. While searching for a model line and one or a plurality of sub-model lines set inside or outside the model line set to indicate a specific feature amount based on the brightness, a search is performed on these model lines and sub-model lines. ,
The feature amount based on the different brightness at the coincidence point and the non-coincidence point with the contour is calculated, and the model line position at which this feature amount is closest to the specific feature amount in the specific positional relationship is determined as the position of the object. A position recognition method characterized in that the position is detected as 3. The feature quantity is a ratio of the number of brightness positions corresponding to the contour line on the model line to the number of brightness positions corresponding to the contour line on the sub model line. Item 3. The position recognition method according to Item 2. 4. The position recognition method according to claim 2, wherein the feature amount is a ratio between brightness on the model line and brightness on the sub model line. 5. The method according to claim 2, wherein a blur of the object or its outline on the image before and after the conversion is detected, and the sub model line is moved to a position outside the range in which the blur is generated or is newly set and used. The position recognition method according to any one of claims 1 to 4. 6. The position recognition method according to claim 1, wherein the image conversion is performed by performing an edge enhancement operation and then performing a density conversion on the brightness on the model line and the sub model line. 7. The position recognition method according to claim 1, wherein the same model line or the sub model line is used for detecting the positions of a plurality of different objects. 8. An image conversion means for converting an image obtained by capturing an object so as to extract a contour of the object, and specifying a contour of the object in a specific positional relationship with respect to the contour of the object by the image conversion means. While searching with a model line set to indicate a specific feature based on their brightness, having a matching point and a non-matching point, the matching point between the model line and the contour line on the model line at each search position A search unit that obtains a different brightness at the mismatch point and calculates a feature based on the obtained brightness, and the feature calculated as a result of the search is closest to the specific feature in the specific positional relationship. A position recognition device comprising: a position detection unit that detects a model line position as a position of an object. 9. An image conversion means for converting an image obtained by capturing an object so as to extract an outline of the object, and a specific positional relationship with respect to the outline of the object on an image converted by the image conversion means. While searching with a model line and one or a plurality of sub-model lines set inside or outside the model line and having a specific coincidence point and a non-coincidence point and indicating a specific feature amount based on their brightness A search unit that obtains different brightnesses on a model line and a sub model line at each search position at a point of coincidence with a contour and a point of non-coincidence, and calculates a difference in the obtained brightness as a feature amount; Position detecting means for detecting, as the position of the object, a model line position at which the calculated characteristic amount is closest to the characteristic amount due to a specific brightness difference in the specific positional relationship. A position recognition device characterized by the above-mentioned. 10. The feature quantity is a ratio of the number of brightness positions corresponding to the contour line on the model line to the number of brightness positions corresponding to the contour line on the sub model line. Item 10. The position recognition device according to Item 9. 11. The position recognition apparatus according to claim 9, wherein the feature amount is a ratio between brightness on the model line and brightness on the sub model line. 12. A blur detecting means for detecting blur of an object or an outline on an image before and after conversion, and moving or newly moving a sub-model line outside a range in which blur detected by the blur detecting means has occurred. 10. Model line changing means for changing a sub model line so as to set and use the set model line.
12. The position recognition device according to any one of items 11 to 11. 13. The image conversion means includes edge enhancement means for performing an edge enhancement operation, and density conversion means for performing density conversion of brightness on the model line and the sub-model line.
The position recognition device according to claim 1. 14. The position recognition apparatus according to claim 9, wherein the search means uses the same model line to search for a plurality of different contour lines.