JPH0981748A - Method for discriminating outline shape and automatic sorting device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To sort an inspecting object according to the thickness, weight, etc., in addition of its outline shape. SOLUTION: Light is projected on the inspecting object, mounted on the mount surface 8 of an inspection stage from a light source 10 and its projection image is picked up by a camera 9. A processor 15 processes the picked-up projection image to represent its outline coordinates by an orthogonal coordinate system, and converts them to a cylindrical coordinate system having its origin at the center of gravity of the projection image so that θ becomes constant. A normalization correlative value between the variation curve of θto (r) at the converted coordinates and the variation curve of a master which is previously stored is calculated to sort the shape. At the same time, the thickness and weight are measured by a thickness measuring sensor 11 and a scale 12 and compared with those of the master to control a sorting mechanism 4 which classifies the inspecting object by shapes according to the sorting result.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method for identifying the contour shape of an opaque object and an apparatus for selecting an object by this identification.

[0002]

2. Description of the Related Art A chain coding method and a geometrical feature extraction method are known as means for identifying the contour shape of an opaque object. The chain coding method is a method in which the positional relationship between pixels is expressed by a predetermined direction index from the digitized image information, and the contour shape is expressed by a chain of direction indexes connected in a chain (Patent Publication 4- (See Publication No. 17079). The geometric feature extraction method is a method of measuring the number and angles of vertices from the contour information of a digitized image, and is used to identify a clear shape such as a triangle or a quadrangle.

Further, as another contour shape identifying technique, a feature amount such as an area, an aspect ratio, and an outer peripheral length is obtained from a digitized image of an object to be examined, and whether or not it agrees with a preset value. There was something to judge. Further, as a technique for identifying not by the contour but by the entire shape, as shown in FIG. 11, a binarized digital image 30 of the test object and a master image are used.
There was a pattern matching method that took the difference from 31.

[0004]

However, in the above shape identification technique, it is necessary to select the feature extraction means according to the shape to be identified. Since the shape of the feature extraction means once set is limited, it is not easy to add a new shape. In particular, the chain coding method recognizes that even if the shapes are similar, the shapes are different because the number of data changes when the size of the object changes. In addition, the geometric feature extraction method cannot be applied to shapes that do not have clear vertices.

Also in the pattern matching method, there is a problem that when the size of the object is changed, the shape is recognized differently, and when the object is displaced in the rotational direction, the object cannot be identified. On the other hand, although there is shape identification by image processing using transmitted light, it is impossible to perform selection considering factors such as thickness and weight. In the end, there is no automatic sorting device that can discriminate any shape and consider the thickness, weight, and material, and such sorting must be done manually.

[0006]

The present invention has been made in order to solve the above-mentioned problems, and a projected image of an object to be inspected is image-processed to be represented by a row of pixels, and the contour coordinates are expressed in a rectangular coordinate system. After all measurements, interpolate and convert into a cylindrical coordinate system with the center of gravity of the projected image as the origin so that the interval of θ becomes constant, and the change curve of r with respect to θ and the change of the master stored through similar processing in advance The feature is that the shape of the object to be inspected is identified by obtaining a normalized correlation value with the curve. By converting the obtained contour data into a cylindrical coordinate system so that the interval of θ becomes constant, the change in the number of data due to the difference in the dimensions of the test object is eliminated.

Further, in order to identify a subtle difference in shape such as a square and a rhombus, a vector from one point to an adjacent point in the contour coordinates represented by the cylindrical coordinate system is sequentially obtained and has the same direction component. The change curve of the number of data with respect to the angle of the vector is obtained by counting the number of data, and the normalized correlation value between this change curve and the change curve of the master stored through the similar processing in advance is obtained to obtain the test object. It is characterized by identifying the shape of. Shape recognition may be performed by combining the above two methods.

Further, the above-mentioned change curve obtained from the projected image of the test object is shifted one by one by the number of data, and the normalized correlation value with the change curve of the master is calculated every time, and the maximum correlation is calculated. It is preferable to recognize the shape of the test object from the value.
Thereby, the shape recognition can be performed without being affected by the rotation of the test object. Then, when the above-mentioned change curve shows a constant value, the shape of the test object is recognized as a circle. Specifically, before calculating the correlation value, the variance value of the change curve data is calculated, and when the variance value is small, the correlation value is not calculated as singular data, and division by 0 is prevented.

An automatic sorting apparatus using such a method is characterized by the following respective configurations. An inspection stage on which the inspection object is placed, a light source that projects light from the back side of the inspection stage onto the inspection object, a camera that is located on the front surface side of the inspection stage and that captures a projected image of the inspection object, and image processing of the captured projection image Then, the means for recognizing the contour coordinates in the Cartesian coordinate system, the contour coordinates expressed in the Cartesian coordinate system are interpolated so that θ becomes constant in the cylindrical coordinate system whose origin is the center of gravity of the projected image.
A means for converting, a means for calculating a normalized correlation value between a change curve of θ for r in interpolated / converted coordinates and a master change curve stored in advance by the same processing, and a test from the normalized correlation value A sorting mechanism that determines the shape of an object and classifies the object to be inspected according to the shape.

An inspection stage on which an object to be inspected is mounted, a light source for projecting light from the back side of the inspection stage onto the object to be inspected, a camera which is located on the front surface side of the inspection stage and takes a projected image of the object to be inspected
A means for processing the captured projection image by image processing to express its contour coordinates in a rectangular coordinate system, and interpolating the contour coordinates expressed in a rectangular coordinate system so that θ is constant to the cylindrical coordinates whose origin is the center of gravity of the projected image. -Means for converting, means for sequentially obtaining vectors from a certain point to an adjacent point in the interpolated / transformed contour coordinates, and counting the number of data having the same direction component to obtain a change curve of the number of data with respect to the angle of the vector , A means for calculating a normalized correlation value between the change curve and a master change curve which is similarly processed and stored in advance, and the shape of the object to be inspected is determined from the normalized correlation value, and the object is analyzed for each shape. It has a sorting mechanism to classify into.

In the above apparatus, the inspection stage should be provided with at least one of a sensor for measuring the thickness of the test object and a scale. Thereby, in addition to the contour shape, the thickness and weight of the test object can be used as the selection criterion. Further, in addition to the above configuration, it must be provided with a means for calculating the area of the image of the test object. Since the thickness and weight can be known by the thickness measuring sensor and the scale, the specific gravity can be calculated if the area is known, and the area and the specific gravity can also be used as selection criteria.

[0012]

EXAMPLES The present invention will be described below based on examples in which throw-away chips (hereinafter referred to as TA) are selected.
FIG. 1 is a block diagram of the device of the present invention. As shown in the figure, it comprises an inspection stage 1, a TA supply mechanism 2 and a sorting mechanism 4 for sorting TA into a sorting box 3. The supply mechanism 2 and the sorting mechanism 4 are both biaxial robots. The former grips one of many TAs on the parts feeder 5 with the chuck 6 and conveys it onto the inspection stage, and the latter grips the TA on the inspection stage with the chuck 7 and sorts the classification boxes 3 based on the inspection results. It is put in the predetermined area of. The number of areas in the classification box 3 may be set appropriately in accordance with the number of selection patterns. It is not necessary to place the TAs on the inspection stage in the same direction, and the TAs may simply be placed on the inspection stage 1.

The details of the inspection stage 1 are shown in FIG. The inspection stage 1 has an ITV camera 9 above the mounting surface 8 of the TA, a halogen light source 10 below and a thickness measuring sensor 11 on both sides, and the table portion functions as a scale 12. The light source 10 projects diffused light onto the test object (TA) on the mounting surface, and the camera 9 captures a projected image of the TA. Further, as shown in FIG. 3, the thickness measuring sensor 11 is composed of a light projector 13 and a light receiver 14 which are arranged to face each other. When there is nothing on the mounting surface, all the laser slit light emitted from the projector 13 and parallel to the mounting surface 8 reaches the light receiver 14 (see FIG. 3A). On the other hand, if TA is present on the mounting surface, a part of the slit light is blocked, so that the thickness of TA can be recognized from the distance where the light cannot be received (FIG. 3B).
reference). In this way, the TA is imaged from above the mounting surface 8 and the thickness is measured from the side of the mounting surface 8, so that both operations do not interfere with each other. The scale 11 calculates the weight from the difference between the measured values before and after mounting the test object. At the inspection stage 1, the TA is imaged, the thickness is measured, and the weight is measured, and all the results are sent to the processing device 15 (FIG. 2). Processor
15 sorts TAs according to the procedure described later from the sent data and sends the data to the sorting mechanism control unit 16 to control the sorting mechanism 4.

The structure of the processing device 15 is shown in FIG. The image captured by the ITV camera 9 is digitized by the A / D converter 17 and stored in the memory 18. The memory 18 is connected to the CPU 19 and extracts the outline coordinates of the image digitized by the CPU 19. The contents of the memory 18 become an analog signal by the D / A converter 20 and can be confirmed by the monitor 21. In the processor 15, the image data is subjected to predetermined arithmetic processing to select TA. That is, a θ-r change curve (described later) of the master pattern (shape that serves as a reference for selection) is registered, this change curve is also obtained for the test object, and the shape of the test object is determined from the correlation value of both change curves. Which master pattern corresponds to is determined. Also, if necessary, a Vθ-n change curve (described later)
In the same manner, the correlation value is obtained, and the shape is distinguished from the result. In that case, you may add thickness, weight, etc. as a selection criterion. When selecting based on thickness and weight,
The thickness, weight, etc. of the test object are measured and compared with the master data to judge whether or not they match.

<Registration of Master> Here, a case where selection based on the θ-r change curve and selection based on the Vθ-n change curve are used together will be described as an example. First, the shape to be registered (master may be TA itself) is placed on the inspection stage, and the classification number is designated. The classification number corresponds to each master pattern, and if it is determined that the pattern has the same shape as the master pattern, a signal of the corresponding classification number is sent to the sorting mechanism control unit. The master 9 is imaged by the camera 9, the image sent to the processing device 15 is represented by a rectangular coordinate system, and its contour coordinates are extracted. In this example, a square master (TA) is imaged,
The contour data is represented by a plurality of pixels as shown in FIG.

Next, the contour coordinate data is converted into cylindrical coordinates with the center of gravity C of the projected image as the origin. The conversion is performed so that the cylindrical coordinate θ is constant (see FIG. 5B). That is, the distance from the center of gravity C to a pixel with contour coordinates is r _0, and the distance from the center of gravity C to the contour pixel for each angle θ is r ₁ , r
_The contour coordinates are represented as ₂ , r ₃ . θ may be appropriately set in consideration of the sorting accuracy and the calculation speed. This makes it possible to keep the number of data constant regardless of the size of the test object. The change of r with respect to θ in this cylindrical coordinate (θ−
The graph of (r change curve) is shown in FIG. 6 (A). Further, as shown in FIG. 6D, the vector direction Vθ from a certain pixel to an adjacent pixel in the contour coordinates of the cylindrical coordinate system is sequentially obtained, and the number of data having the same direction component is counted to calculate the vector angle. A change in the number n of data with respect to Vθ (Vθ-n change curve) is obtained (see FIG. 6C). The data of both change curves described above is stored in the memory 18 as a master pattern.

In addition, as the master data, the thickness, weight, area, and specific gravity of the master are registered. The thickness and weight are measured by the thickness measuring sensor 11 and the scale 12. The area is calculated from the orthogonal coordinate data of the projected image. The specific gravity may be obtained by obtaining the volume from the area and the thickness and dividing this by the weight. When registering multiple masters, the above procedure is repeated for each master. The number of registered masters is possible as long as the amount of memory allows.

<Selection of TA> When selecting TA, the θ-r change curve and the Vθ-r change curve of the test object are obtained in the same manner as above. Then, the normalized correlation value with the change curve of the master is calculated based on Expression 1. Since Equation 1 normalizes the change curve, an object similar to the master is recognized as a similar change curve. If the correlation value is a value close to 1, the shape of the test object is recognized as the same as or similar to the master, and as it approaches 0, it is recognized as a completely different shape. Even if the shape of the object to be inspected is not registered as the master pattern, the correlation value with the closest master pattern is smaller than 1, and therefore it can be distinguished as “other”.
Although the threshold value for identifying as “other” varies depending on the object to be inspected, a good result was obtained in this TA selection of about 0.8.

[0019]

[Equation 1]

For example, if an object to be inspected that is congruent with the master pattern is placed on the inspection stage in the same direction as the master, the θ-r and Vθ-n change curves of the object to be inspected will be the same as those of the master, and the correlation value will be the same. Becomes 1. Further, when an object similar to the master pattern is placed on the inspection stage in the same direction, the θ-r and Vθ-n change curves are as shown in FIG. 7. In the figure, the solid line is the change curve of the master pattern, and the broken line is the change curve of the test object. Since the θ-r change curve is normalized to obtain the correlation value, the curves (solid line and broken line) as shown in FIG. 7A are recognized as the same curve. The determination of the contour shape may be performed only by the θ-r change curve,
This may be combined with the identification based on the Vθ-n change curve.
It may be appropriately selected according to the sorting accuracy. In particular, Vθ−
Since n-curve identification can consider subtle differences in angles, rectangles and parallelograms, squares and rhombuses such as θ-r
Shapes that are difficult to identify with change curves can be clearly distinguished.

By the way, since the inspection object is placed on the inspection stage in the same direction, even if the master and the inspection object have the same shape, the phases of the respective change curves do not match first and the correlation value is low. Therefore, the data of the change curve of the master or the test object is shifted by one data by the number of data, and the correlation value with the change curve of the master is calculated each time. Then, the shape is determined from the maximum correlation value. That is, when the master pattern and the test object have the same shape, the maximum correlation value becomes 1 when the data of the change curves are sequentially shifted and the phases of both curves match. From this fact, the shape is determined regardless of the orientation of the object.

For example, for a master of an isosceles triangle,
When inspecting a test object which is conjoint and has a different direction, the both θ-r change curves are out of phase and do not match as shown in FIG. 8 (A). However, if one of the change curves is sequentially shifted, the correlation value is calculated for each time, and the correlation value is graphed, it becomes as shown in FIG. 7B, and the maximum value becomes 1, so it is the same as the master. Can recognize sex. At this time, it is possible to know how much the test object is rotating with respect to the master from the shift amount of the data.

When the shape of the object to be inspected or the master is circular, as shown in FIG. 9, the θ-r change curve (FIG. 9B) and V
Both θ-n change curves (C in the same figure) are constant values. In that case, the denominator of Equation 1 becomes 0, and a normal correlation value cannot be calculated. Therefore, the variance of the change curve data is calculated prior to the calculation of the correlation value, and when the variance value is small (for example, 0), the correlation value is not calculated and the shape of the test object may be determined to be circular. The change curve shows a constant value only when the test object is circular.

The procedure described above is arranged in a flow chart as shown in FIG. First, "register master pattern" is performed, and "addition" is performed as necessary. next,
As the “setting of identification parameter”, θ-r change curve and V
One or both of the θ-n change curves are selected, and at least one of thickness, weight, area, and specific gravity is selected as necessary. The object is placed on the inspection stage and each parameter is measured. The variance value is calculated from the data of the θ-r change curve and / or the Vθ-n change curve, and if it is 0, the shape of the test object is determined to be circular. If the variance value is not 0, the θ-r change curve and the Vθ-n change curve of the test object are obtained, and the correlation value with the same change curve of the master pattern is calculated. Subsequently, "data shift" is performed on each change curve of the test object and the master, and the maximum correlation value between the two is calculated. When there are a plurality of masters, a series of correlation value calculations are similarly performed for all masters. The master pattern may be changed as needed. Then, the test object is selected based on the obtained correlation value data, thickness, weight, area, and specific gravity. In the actual sorting work,
The selection time can be shortened by performing these series of calculations during the settling time of the balance. With the automatic sorting apparatus developed this time, a TA of 50 mm square or less was used as the test object, and a sorting ability of 3 seconds / piece was obtained when 30 patterns were sorted.

[0025]

As described above, according to the present invention, the shape of an object having an arbitrary shape can be identified simply by placing the object on the inspection stage without complicated setting. can do. The registration / addition of the master pattern only needs to be placed on the inspection stage in the desired shape. In addition, it is possible to select by taking into consideration criteria such as thickness, weight and specific gravity. Therefore, TA, bolts, nuts,
It is effective when used for selection inspection of machine parts with various shapes such as gears. In particular, by appropriately adding selection criteria such as thickness, weight, area, and specific gravity in addition to the contour shape, various selections from rough selection to fine selection can be performed.

[Brief description of drawings]

FIG. 1 is a block diagram of a sorting apparatus of the present invention.

FIG. 2 is a configuration diagram of an inspection stage of the device of the present invention.

3A and 3B are explanatory views showing a thickness measuring sensor, where FIG. 3A shows a state without a test object, and FIG. 3B shows a state with a test object.

FIG. 4 is an explanatory diagram of a processing device in the device of the present invention.

5A is an explanatory diagram showing extraction of contour coordinates in rectangular coordinates, and FIG. 5B is an explanatory diagram showing interpolation / conversion from a rectangular coordinate system to a cylindrical coordinate system.

6A is a graph showing a θ-r change curve when a square test object is measured, FIG. 6B is an explanatory diagram of θ and r in a cylindrical coordinate system, and FIG. 6C is a square test object. A graph showing a Vθ-n change curve when an object is measured,
(D) is an explanatory view of Vθ in a cylindrical coordinate system.

7A and 7B are graphs showing the results of measuring an object similar to the master, in which FIG. 7A shows a θ-r change curve and FIG. 7B shows a Vθ-n change curve.

FIG. 8A is an explanatory diagram showing that the phase of the θ-r change curve is different when measuring a congruent isosceles triangle test object whose orientation is different from that of the master; and FIG. 8B is a shift curve. The graph which shows the change of the correlation value at the time of doing.

9A and 9B are explanatory diagrams of measurement conditions when the test object is circular, in which FIG. 9A shows a contour shape of the test object, FIG. 9B shows a θ-r change curve, and FIG. 9C shows a Vθ-n curve. .

FIG. 10 is a flowchart showing a selection processing procedure of the method of the present invention.

FIG. 11 is an explanatory diagram of a pattern matching method.

[Explanation of symbols]

1 Inspection stage 2 Supply mechanism 3 Sorting box 4 Sorting mechanism 5 Parts feeder 6 Chuck 7 Chuck 8
Mounting surface 9 ITV camera 10 Light source 11 Thickness measurement sensor 12
Scale 13 Emitter 14 Light receiver 15 Processor 16 Sorting mechanism controller 17
A / D converter 18 Memory 19 CPU 20 D / A converter 21 Monitor 30 Image of test object 31 Master image

Claims (12)

[Claims] 1. A projection image of a test object is image-processed and expressed by a pixel row, and after measuring all of the contour coordinates in a rectangular coordinate system, an interval of θ is set in a cylindrical coordinate system whose origin is the center of gravity of the projection image. Is interpolated / converted to be constant, and the shape of the object to be inspected is identified by obtaining a normalized correlation value between the change curve of r with respect to θ and the change curve of the master stored in advance and similarly stored. A method for identifying a characteristic contour shape. 2. A projected image of an object to be inspected is image-processed and expressed by a pixel array, all of the contour coordinates are measured in a rectangular coordinate system, and then the interval of θ is set in a cylindrical coordinate system whose origin is the center of gravity of the projected image. Is interpolated and converted so that the vector becomes constant, the vectors from a certain point in the contour coordinates converted into the cylindrical coordinate system to the adjacent points are sequentially obtained, and the number of data having the same direction component is counted A contour shape characterized by identifying a shape of an object to be examined by obtaining a change curve of the number of data and obtaining a normalized correlation value between this change curve and a change curve of a master which is similarly obtained and stored in advance. Identification method. 3. A method for identifying a contour shape, characterized by combining the method according to claim 1 and the method according to claim 2. 4. The change curve obtained from the projected image of the test object is 1
4. A method of shifting a point-by-point number by the number of data, calculating a normalized correlation value with the change curve of the master each time, and recognizing the shape of the test object from the maximum correlation value.
A method for identifying a contour shape according to any one of 1. 5. When the change curve shows a constant value, the shape of the test object is recognized as a circle.
4. The contour shape identification method according to any one of 4 above. 6. An inspection stage on which an inspection object is placed, a light source for projecting light from the back surface side of the inspection stage onto the inspection object, a camera which is located on the front surface side of the inspection stage, and which captures a projected image of the inspection object, A means for recognizing the contour coordinates in a rectangular coordinate system by image processing the projected image. The contour coordinates expressed in the rectangular coordinate system are interpolated so that θ is constant in a cylindrical coordinate system whose origin is the center of gravity of the projected image. A means for converting, a means for calculating a normalized correlation value between the change curve of θ with respect to r in the interpolated / converted coordinates and the change curve of the master which is similarly obtained in advance and stored, and to be tested from the normalized correlation value An automatic sorting apparatus, comprising: a sorting mechanism that determines a shape of an object and classifies an object to be inspected according to the shape. 7. An inspection stage on which an inspection object is placed, a light source for projecting light from the back side of the inspection stage onto the inspection object, a camera which is located on the front surface side of the inspection stage and takes a projected image of the inspection object, A means for image processing the projected image and expressing the contour coordinates in a Cartesian coordinate system. The contour coordinates expressed in a Cartesian coordinate system are interpolated / converted into cylindrical coordinates with the center of gravity of the projected image as the origin so that θ is constant. Means for sequentially obtaining a vector from a certain point to an adjacent point in the interpolated / converted contour coordinates, and counting the number of data having the same direction component to obtain a variation curve of the number of data with respect to the angle of the vector, A means for calculating a normalized correlation value between a change curve and a master change curve that is similarly obtained and stored in advance, and a classification that determines the shape of the test object from the normalized correlation value and classifies the test object for each shape. mechanism Automatic sorting apparatus characterized by comprising a. 8. The automatic sorting apparatus according to claim 6 or 7, further comprising a supply mechanism for mounting an inspection object on the inspection stage. 9. The automatic sorting apparatus according to claim 6, wherein the inspection stage is provided with a sensor for measuring the thickness of the object to be inspected, and the contour shape and the thickness are used as the criteria for selecting the object to be inspected. 10. The automatic sorting apparatus according to claim 6, wherein a scale of the test object is provided on the inspection stage, and the contour shape and the weight are used as the selection criteria of the test object. 11. The inspection stage is provided with a sensor for measuring the thickness of the object to be inspected and a scale, and the contour shape, the thickness and the weight are used as criteria for selecting the object to be inspected. Automatic sorting device. 12. A means for calculating the area from the coordinates of a projected image, a sensor for measuring the thickness of a test object provided on an inspection stage, a scale provided on the test stage, and an area / thickness of the test object. 8. The automatic sorting apparatus according to claim 6, further comprising means for calculating a specific gravity from the weight, wherein the difference in the specific gravity is used as a criterion for sorting the test object.