JPH02176876A - Graphic recognition device - Google Patents

Abstract

PURPOSE:To emphasize an input image more sharply and to eliminate a noise by performing contour extraction by using a negative Laplacian arithmetic filter following a Sobel arithmetic filter. CONSTITUTION:An original picture in a frame memory 10 passes through a filter at a filter processing part 42 in a filter 4, and the absolute value of an obtained value is written on a frame memory 11. Next, the original picture in the frame memory 10 passes through the filter, and the absolute value of the obtained value is written on a frame memory 12. And finally, the contents of the frame memories 11 and 12, after passing through an adder 5, then, being added, passes through a Laplacian arithmetic processing part 50, and a Laplacian arithmetic operation is applied on it, then, it is written on the frame memory 10. Thereby, the input picture is inputted in the frame memory 10 as it is originally, and it is emphasized by a Sobel arithmetic operation, and it is substituted as data having only a contour part with more change of density in a form where it is much more differentiated by the Laplacian arithmetic operation.

Description

[Detailed description of the invention] [Summary] Regarding a figure recognition device that performs L2 recognition of a circular figure using a two-dimensional wave equation, the purpose is to make the final peak obtained sharp. The outline of the input image of the object is extracted by a Sobel operation filter, converted into a two-dimensional gray value, and inputted to the calculation means, and the center of the circular object is determined by the peak value of the amplitude in the solution of the two-dimensional wave equation with the gray value as the initial value. A figure recognition device that recognizes a position and a radius in L2 is configured to extract contours using a negative Laplacian calculation filter following a cough Sobel calculation filter.

[Industrial application field]

The present invention relates to a figure recognition device that recognizes circular figures using a two-dimensional wave equation.

Circular shapes play a large role in the visual recognition of objects in the mechanical industry.Most of the work done by lathes and milling machines is to form circles, and bolts and rivets are fastened in round holes. In addition, the wiring terminals of printed circuit boards widely used for mounting electronic components are often ring-shaped.Since the wiring terminals are formed by printed wiring, the position of each terminal is known, but there may be some misalignment. Since this is unavoidable, it is desirable to determine the center position of the terminal portion during automatic bonding to ensure accurate bonding.

In addition, in production facilities that have introduced robots, the robots pick up parts made and transported in upstream processes one by one and place them on processing machines, or move them to conveyors in downstream processes. However, in this case, the parts position,
It is necessary to form the shape correctly, and for this purpose it is necessary to accurately measure the center position coordinates and radius of the circle.

[Conventional technology]

Various methods for recognizing circular objects have been proposed, but most conventional methods extract the contour and calculate the center of gravity and circumference, so when the contour line is interrupted, it cannot be traced as a closed loop, Sometimes the circumference cannot be calculated. If this discontinuity is minute, interpolation techniques are known to fill in the gap, but if large parts are missing, intercoupling is often difficult. This drawback occurs when processing gray scale images using reflected light.
This is particularly problematic.

Therefore, L2 that is not distracted by background, surface patterns, and noise.
One example of this method is a method of extracting skeleton lines by generating waves from around a 0 image, which has been pointed out to be effective in processing and macroscopic methods.

Using a similar idea, we compare the circumference of a circular object to waves on the water surface.
Japanese Patent Application Laid-Open No. 58-51387 discloses a figure recognition method for determining the center position by utilizing the fact that waves propagated from the outer periphery gather at the center after a certain period of time and form a peak.

This figure recognition method uses a wave equation to recognize circular objects. Recognizing a circular drawing using the wave equation is done as follows.0 For example, when water is filled in a circular basin and a shock is applied to greenery, waves generated from the surroundings propagate toward the center, and after a certain period of time, they peak at the center. will occur. Therefore, if we regard the area around the circular object as a part surrounded by a circle on the water surface and simulate the movement of waves inside the calculator, we get the following:
Similarly, a peak should occur at the center. If the position of the peak is known, the center position of the circle can be found, and its radius can be determined from the time it takes to reach the peak position.

The strength of this recognition method is that it allows some circles to be missing or multiple circles to overlap each other. This is because the reason why the peak occurs is that the wave forces that occur at the same time on the circumference -E all strengthen each other in the same phase, but even if a part of the circle is missing, the remaining arc is sufficient. This is because if the wave length is long enough, the waves generated there will be in the same phase and strengthen each other. As for the patterns and noise on the surface, unless they are circular, the phase of the waves generated will be random and will not have a large peak, and will be independent of the waves generated from the circumference and will not interfere with them.

However, as shown in FIG. 9, for example, when a circular figure to be recognized has different shades of black and white, the circular figure can be recognized using the conventional method using the wave equation.
If you try to recognize it with a two-way recognition device, the waves of the white part and the black part will weaken each other at the center of the circle.
There will be no amplitude peak at the center of the circle.

Furthermore, if the image of a circular object does not have sufficient contrast against the background, the peak value that occurs at the center position of the circle is small, and therefore the center of the circle may not be accurately recognized.

Therefore, the present inventor proposed a device that can accurately recognize a circular figure even when the shading changes in the circular figure part or when the contrast with the background is small. It was already disclosed in the issue.

In this device, instead of using the input image of the circular object itself as the initial value of the two-dimensional wave equation as described above, the contour of the circular object is extracted and emphasized by applying a Sobel calculation filter to the human image. Processing is performed to address changes in shading of circular objects or lack of contrast with the background.

[Problem to be solved by the invention]

If we apply the input image extracted and enhanced according to the above patent application No. 63-2131.58 as the initial value of the two-dimensional wave equation, the image data to be subjected to the wave calculation will be All of them are positive, so there are cases where only a gentle peak is obtained, and in such cases there is a problem that it is not possible to accurately recognize circular objects.

Therefore, the present invention extracts the contour of a human-powered image of a circular object using a Sobel calculation filter, converts it into a two-dimensional gray value, and inputs it to a calculation means, and the amplitude peak value in a solution by a two-dimensional wave equation that uses the gray value as an initial value. It is an object of the present invention to provide a figure recognition device that recognizes the center position and radius of a circular object so that the peak finally obtained is sharp.

[Means to solve the problem]

In order to achieve the above object, the figure Plti according to the present invention
In this apparatus, as shown in principle in FIG. 1, a Sobel calculation filter 11 is followed by a negative Laplacian calculation filter 12 for contour extraction.

[For production]

In the present invention, the input image waveform when Sobel calculation is performed is as shown in FIG. 2(a), but when it is further passed through negative Laplacian calculation filter 12, differential calculation is performed As a result, a negative portion is generated in the input image waveform as shown in FIG. 2 Φ), so that the outline is further emphasized.

〔Example〕

Hereinafter, the present invention will be described in detail with reference to the drawings. Third
The figure is a block diagram showing a schematic configuration of an image processing processor that implements a figure recognition device as an embodiment of the present invention.

In the figure, 1° to 1 are frame memories, and each frame memory is connected to an 8-bit video bus 3 via a bidirectional bus buffer 2. 4 is a spatial filter, which includes a line buffer 41.3×3 coefficients, product-sum/filter processing unit 42, and absolute value/scaling circuit 43.
Contains.

5 is an adder circuit, and this adder circuit 5 and the spatial filter 4
As shown by the broken line, they are connected via a Vibration line via an internal bus to form a Sobel arithmetic filter. This adder circuit 5 is provided with a Laplacian calculation processing section 50 as a Brassian calculation filter. Reference numeral 6 denotes a conversion circuit that A/D converts and captures an image signal output from a camera (not shown), or performs D/A conversion for monitor output.

The operation of this embodiment device will be explained below.

In conventional figure recognition devices, the gradation value itself of the input image was used as the initial value when solving the wave equation, but in the present invention, the first JIJI (a) of the wave equation is not the human image itself, but the outline is extracted and Use the emphasis.

Sobel operation and Laplacian operation are used as a method for emphasizing the contour. This is generally provided in standard image processors, and this makes it possible to easily realize this 8I function even in the case of the device of this embodiment, which can execute calculations at high speed (for example, 33m5). It is possible.

That is, the Sobel operation and the Laplacian operation can be schematically illustrated as shown in FIG.
The original image in the filter 4 is processed by the filter processing unit 42 in the filter 4.
Then, write the absolute value of the obtained value to frame memory 1.

Next, pass the original picture in the frame memory l again through the
Write the absolute value of the obtained value in frame memo 1.

Finally, the contents of frame memories h and 1g are added to the adder 5.
After addition, Laplacian calculation is performed through Laplacian calculation processing section 50, and frame memo 1 is added.
Month. write to.

As a result, as shown in the density cross-sectional diagram of FIG. 4(B), the frame memory 10 initially contains the input image as it is, but it is enhanced by the Sobel operation and further differentiated by the Laplacian operation. The data is replaced with data only for the contour part, which has a further change in density in shape.

The contour data of the circular object obtained in this way is processed using a wave equation in the following manner, thereby determining the center and radius of the circular object.

First, the brightness of a grayscale image is regarded as a three-variable function of the horizontal and vertical coordinates x, y and the time, and this is defined as A, (x, y,
t) and input the l-frame image A at time T0.
The following two-dimensional wave equation is solved using (x, y, to) as initial values.

tZ 9x" 9y" However, a is the speed of the wave.

The boundary condition of this equation is always A(χ, y, t)=0 on the periphery of the image frame (square).

Physically, this is the same as the equation that describes the state of an elastic membrane fixed around a square.

In order to convert equation (1) into a difference equation and solve it sequentially, the following new variable y (x, y, t) = -A (x, y, t)
Introduce at and make it a first-order simultaneous equation.

That is, when equation (1) is rewritten using V (x, y, t), it becomes as follows.

−V (x, y, t)=a”ΔA(x, y, L)9 X
, "9 y" - A (x, y, t) = V (x, 3
', t)ru.

V (x, y, t+δt/2)= V (x, y, t-δt/2)+ at δtΔA(x, y, t,) A(x, y, L+51)= A (x, y, L)+ δtv (x, y, t+δt/2) ・−−−−−・−(3) The algorithm for solving equation (3) is given below.

Step 1: V-0, A-A(0) t-0 In this example, a2δt-
The calculation is performed as i, δt-1/8, and the procedure at that time is shown in FIG.

The processing in this embodiment uses at least two frame memories. Namely, they are frame memory IJA and frame memory V, and these frame memories are physically connected to frame memory 1-. Corresponds to 1+. The frame memory A is made to hold the two-dimensional gradation value of the image after a predetermined time, and the frame memory 2 is made to hold the time differential value of the two-dimensional gradation value.

In step Sl, the two-dimensional fA light value A(xy, 0) of the original image
is written to frame memory A and the frame memory ■
Write O to , and further set the counter n indicating the number of approximations to 1.
Set to .

In step S2, find the Laplacian ΔA of the image held in the frame memory A, and combine this with the frame memory ■
The result of adding the contents of is written to the frame memory ■ again.

Next, in step S3, the contents of the frame memory ■ are set to 1.
This value is multiplied by a predetermined number less than 1/8, for example, and this value is added to the contents of frame memory A, and the result is written to frame memory A again.

In step S4, values of addresses x, y, and 0 in the frame memory that are greater than a predetermined darkness value S are detected, and these are written into a peak table as shown in FIG.

In step S5, the value of the counter n is incremented by one, and in step S6, it is determined whether n is greater than or equal to R/2-7'T pixels. Here, R is the maximum radius of the circle to be found, and it is assumed that the wave advances by 2f1 pixels in one short cycle. In the case of a negative determination in step S6, steps S2 to S5 are repeated, and in the case of an affirmative determination, the process ends.

Through this process, frame memory A with a predetermined darkness value S or more
The center position of the circular object is determined from the x, y address of , and the radius of the circular object is determined from the number of repetitions.

In the flowchart of FIG. 5, step S2 and step S
Step S3 means performing calculations on all pixels of each frame memory, and the time required for each step is one frame period (for example, about 33 srs).
Assuming that the repetition of Step 2 and Step S3 is referred to as one approximate cycle, the frame memory A holds A (x, y, L+nΔt) at the time when the n-th approximate cycle has progressed.

In steps S2 and S3, since it is necessary to perform calculations on the entire contents of the frame memory within one frame period, a high-speed image processor is generally required to execute this processing routine. The peak table in step S4 is as shown in FIG. 6, and from this peak table, the locus lx at the center of the circle,
y and its radius can be easily determined.

The calculation of the wave equation in the embodiment described above can also be realized by configuring as follows.

That is, some types of image processing sensors may not be able to execute the processing shown in FIG. 5 from the viewpoint of processing speed. for example,
Although the Laplacian ΔA is added to the data read from the frame memory V in step S2 and written into the same frame memory V, it is generally difficult to do this within one frame period.

This is because it is necessary to switch between successive reading and writing of memory elements within the sampling period. Each frame memory has 1
It is easier to control and the circuit is simpler if it is always in the successive state or write state within a frame period.

Further, depending on the processor, it may not be possible to apply the Laplacian to one frame memory and add it to another frame memory as in step S2 in FIG. 5. In other words, it may have only a simple frame addition function. In order to execute the above-mentioned processing on a processor having such a simple function, the number of working memories may be increased and the processing procedure shown in FIG. 7 may be performed.

That is, in this modification, in addition to frame memory A and frame memory V, two working memories W, , W
, add. In step Sll, the absolute value of the two-dimensional grayscale value from which the outline of the original image is extracted is written in the frame memory A, and O is substituted in the frame memory V. Further, a counter n indicating the number of close inspections is set to 1 (0).Next, the process proceeds to step Si2, where the Laplacian image ΔA is written into the working frame memory Wl, and the working frame is stored in the memory W.

In step S12, the contents of the frame memory V are added to the data read from the working frame memory W, and the data is written to the working frame memory W, and the data read from the working frame memory W is written to the frame memory ■. Then, in step S13, the data read from the frame memory V is multiplied by 178 and written into the working frame memory W, and the data read from the working frame memory Wl is combined with the data read from the working frame memory Wl. Adding the data read from frame memory A and writing the data in working frame memory Wt to frame memory A, each calculation described above in step 513 requires one frame period, so three frame periods are required in total. .

Step 512 performs the same calculation as step S2 in the flowchart of FIG. 5, and step S13 performs the same calculation as step S3 in FIG.

Next, the process proceeds to step S14, where the values of addresses X, y, and n in the frame memory A that are equal to or greater than a predetermined threshold (13) are written into a peak table as shown in FIG. 6. In step S15, the counter n is incremented by one. , in step S16, it is determined whether n is equal to or greater than R/2JT pixels, and in the case of a negative determination, steps 312 to 515 are repeated, and in the case of an affirmative determination, this processing routine is ended.As described above, this modification In the example, four frame memories are used, and the time required for one approximation is six frames.

In yet another variation, more working memory may be used to perform operations in a shorter time than the embodiment of FIG.

FIG. 8 shows a modification when five frame memories are used.

In FIG. 8, first, in step 521, two-dimensional grayscale values of the original image are written into the frame memory Aeven, and O is substituted into the frame memory veverl.

Further, a counter n indicating the number of times of approximation is set to 1.

In step S22, the Laplacian image ΔAeven of the frame memory Aeven is written to the frame memory W, and the data read from the frame memory W and the data read from the frame memory Veven are written to the frame memory νodd. 1 for the read data
Step S2: Multiply by /8 and write to the frame memory Veven, and add the data read from the frame memory Veven to the data read from the frame memory Aever+ and write to the frame memory Aodd.
In step S2 and step S23, one frame period is required to perform the respective calculations, so two frame periods are required to execute step S22 and step S23, respectively. Step S22 executes an operation equivalent to step S2 in the flowchart of FIG. 5, and step S23 executes an operation equivalent to step S3 of FIG.

Next, the process advances to step S24, where the frame memory ^od
Addresses x, y, and n where d is greater than or equal to a predetermined darkness value S
Write out the values in a peak table as shown in Figure 6.
This means that an odd numbered approximation cycle is executed in steps 322 to S24.

In step S25, the counter n is incremented by one, and the process proceeds to step 326 to step S2B, where the even-numbered nearest cycle is executed.
Exactly the same process is executed by changing ``to'' to ``odd'' and changing ``odd'' to ``even''.

In step S29, the value of the counter n is further incremented by one, and then the process proceeds to step 530, where n is R/2J.
It is determined whether or not the number of pixels is greater than or equal to 10 pixels, and in the case of a negative determination, the process from step 322 is repeatedly executed, and in the case of a positive determination, this processing routine is terminated. In this embodiment, five frame memories are used. By using this, calculations can be executed in a period of 4 frames.

〔Effect of the invention〕

As described above, according to the present invention, a negative Laplacian calculation filter is used following the Sohel calculation filter to perform ring extraction, so that the input image is enhanced more clearly and noise is removed. Therefore, if we apply the two-dimensional wave equation based on this, we will obtain a distinct peak, and therefore we can determine the exact center position and radius of the circular object.

[Brief explanation of the drawing]

Fig. 1 is a principle block diagram of the figure recognition device according to the present invention, Fig. 2 is a waveform diagram for explaining the present invention in detail, and Fig. 3 shows m = embodiment of the figure recognition device according to the present invention. Block diagram; Fig. 4 is a schematic explanatory diagram of a method for extracting and emphasizing the outline of an image of a circular object; Fig. 5 is a flowchart of the circular recognition processing procedure using the wave equation in the embodiment device; Fig. 6 is a peak table FIG. 7 is a flowchart of the figure recognition processing procedure in a modified example, FIG. 8 is a flowchart of the figure recognition processing procedure in another modified example, and FIG. 9 is a diagram explaining the problems of the conventional method. . DESCRIPTION OF SYMBOLS 11... Somol calculation filter, 12... Laplacian calculation filter, 13... Calculation means, 1°~IR... Frame memory, 2... Bidirectional pass bumper, 3... Video bus, 4. ... Spatial filter, 5... Addition circuit. In the figures, the same reference numerals indicate the same or corresponding parts.

Claims (1)

[Claims] The outline of the input image of a circular object is processed by a Sobel calculation filter (1
Calculating means (13) to extract and make two-dimensional gray values by 1)
In a figure recognition device that recognizes the center position and radius of the circular object based on the peak value of the amplitude in the solution according to the two-dimensional wave equation with the grayscale value as the initial value, following the Sobel calculation filter (11), A figure recognition device characterized by extracting contours using a negative Laplacian calculation filter (12).