JPS6029878A - Detector for circular body - Google Patents

Abstract

PURPOSE:To obtain data on an independent circle and an arc by picking up an image of plural bodies whose contours are arcuate or enveloped with an arc, making a raster scan and converting data into binary data, and operating a scanning segment terminal part and the overlap state of segments. CONSTITUTION:An image 12 of the plural circular bodies 11 on a background 10 is picked up while they overlap one another. The video signal obtained by making the raster scan on the image is converted 13 into a binary signal to obtain a picture as shown in a figure 24. This information is written in a picture memory in DMA mode by a picture information input circuit 14 which includes the picture memory and a picture feature extraction part. A connectivity analyzing part 15 detects 16 boundary points on the basis of line segment information (coordinates and length of segment terminal part and whether the segment overlaps an adjacent segment or not) on the binary-coded picture from the feature extraction part, and carries out the trace arithmetic 17 of inner and outer circumferential points, arithmetic 18 of arcuate parts, arithmetic 19 of estimated circles, and output 20 of detection results to constitute body circle detection data. Consequently, the positions, sizes, and number of overlapping circles are detected.

Description

[Detailed description of the invention] [Technical field to which the invention pertains] This invention uses an imaging means such as a TV camera to separate circular objects such as food containers that are randomly arranged, including overlapping, and to separate them into individual objects. The present invention relates to an image processing device that measures the diameter of an object.

[Prior art and its problems]

For example, it is possible to image an individual object using an imaging means such as a TV camera, and identify the object by determining the object's size (area, diameter, outer circumference), etc. from the image signal. It is used to sort objects by identifying whether they are good or bad. In such cases, conventionally it was easy to determine the size of the object from the image signal because each object was separated and independent, but object identification using a TV camera As it is being applied in many fields, it naturally becomes necessary to be able to apply it not only to cases where individual objects are separate and independent, but also to cases where they overlap.

However, when images of overlapping objects are captured using a TV camera, the captured images appear as one object.
I was unable to determine the correct size of each object.

For example, in order to automate the company cafeteria,
The size of the tableware container is associated with the price, and as shown in Figure 1, the container 2 and 3 of the food freely selected by the employee placed on the tray 1 are photographed using the camera 4. It is conceivable to determine the size of the containers 2 and 3 from the image signal and calculate the 1st stage. In such a case, if containers 2 and 3 overlap when viewed from above, the captured image will look like the one shown in Figure 2, making it difficult to determine the respective sizes of containers 2 and 3. Ta. For this purpose, as shown in Figure 1, the TV
The camera 5 is placed below the tray 1, and the lighting 8 is placed above the tray 1.
It is also being considered to obtain an image as shown in FIG. 3 by arranging the thread ends s6 and 7 of the containers 2 and 3 on the tray 1 by arranging the thread ends s6 and 7 of the containers 2 and 3 clearly. In this case, the price can be reduced by matching the size of the thread tail and the price, and since the thread tail will never be equal to 1, the image will look like the one shown in Figure 3. It is easy to determine the size of the end of the thread, as it is always separated.

However, in the case of this device, the end of the thread needs to be clearly visible.

■ When the shadows of other containers that are far apart overlap, ■ When containers are on top of each other, ■ When containers float due to foreign objects and the shadow of the end of the thread is not clearly visible. If the butt cannot be imaged, discrimination becomes impossible.

[Purpose of the invention]

The present invention is directed to an image processing device capable of detecting the number, position, size, etc. of circular objects such as food containers even when circular objects such as food containers are superimposed, and which eliminates the above-mentioned drawbacks. The purpose of this invention is to provide a device that can accurately determine the position and size of a placed circular object and can be applied to a wider range of applications than conventional devices.

[Key points of the invention]

The gist of the present invention is to perform 2-star scanning on images of multiple objects, including superimposed objects, whose contours have a circular arc or an arcuate envelope, and then to binarize the images using a binarizing means. The coordinates of the end of the scanning segment (line segment) of the pattern of the circular object and the adjacent previous scan are converted into an image, and the coordinates of the end of the scanning segment (line segment) of the pattern of the circular object are transferred to the binarized image by a device that detects the position and size of the circular object. Image that detects overlap with i-segment! a feature extracting means, and a delayed analysis means for adding a pattern number for each independent pattern to all scanning segments of the binarized image based on the patterned information from the image feature extracting section; boundary point detection means for detecting the coordinates of the end of the scanning segment having the pattern number as the coordinates of the boundary point of the independent pattern; after separating the boundary point data for each independent pattern, the independent pattern is separated from the outer periphery. Turn counterclockwise (clockwise)
Outer periphery point sequence data, or each independent pattern, consisting of the coordinates of each boundary point arranged in the order of tracking, and a direction code indicating the tracking direction from the boundary point to the adjacent boundary point. The outer peripheral point sequence KJf among the boundary points of
i is tracked along the inner periphery of the pattern in the same way as in the case of the previous hG outer periphery, and one or both of the outer periphery point sequence data and the same inner periphery point sequence data are traced. Outer/inner point tracking calculation means for outputting data, and the value of the direction code in the outer/inner point string data, or the outer/inner point string in the section increasing (decreasing) in the order of the tracking. A convex image that extracts the coordinates of , excludes the concave image part, that is, the connecting part of each object circle, and outputs the convex image part, that is, the outer/inner circumference/convex part point sequence data for each part that seems to correspond to each object circle. The partial separation means sequentially selects outer/inner points constituting the outer/inner periphery/convex point sequence data for each of a plurality of adjacent outer/inner periphery points in order to shorten calculation time. outer/inner circumference/representative point selection means for outputting as an inner circumference representative point, and individual means for calculating and outputting the center coordinates and radii of a plurality of individual arcs that sequentially pass through three adjacent points at the outer/inner circumference representative points. circular arc calculating means; center coordinates of the individual circular arcs;
The same circle/individual arc calculating means separates those whose radius values are close to each other as belonging to the same circle, and the estimated circle derived by applying the least squares method to these individual arcs, that is, the same circle. Estimated circle calculation means calculates and outputs the center coordinates and radius of the estimated circle using the valley data corresponding to each individual arc belonging to the circle; The average value of each of the radii is taken as the center coordinate of the IW fixed circle and the first approximation value of each of the radii. By subtracting the radius of the estimated circle from the distance from the center coordinate of The point is that it has been shortened.

[Embodiments of the invention]

The present invention will be explained below based on FIGS. 4 to 12. FIG. 4 is a block diagram showing the configuration of an embodiment of the detection device of the present invention. In the figure, 10 is a background, 11 is a circular object to be detected, 12 is an imaging means such as a television camera, 22 is a circular object detection device, 13 is a binarization circuit, and 14 is an image memory or image feature extraction unit. An image information input circuit including: 1
5 is a connectivity analysis section, 16 is a boundary point detection section, 17 is an outer/inner circumference point tracking calculation section, and 18 is an arc including a convex image part separation section, an outer/inner circumference/representative point selection section, and an individual arc calculation section. a calculation unit; 19 is a same circle/individual arc calculation unit; an estimated circle calculation unit including a least squares circle calculation unit; 20 is a detection result output unit; 21 is the number of circles;
This is object circle detection data indicating the position and size.

First, before explaining the details, the general operation of FIG. 4 will be explained. A circular object 11 is placed on a background 10, and depending on their shape, when images are taken from above by the imaging means 12, they may overlap. be.

A binarization circuit 1 converts the video signal obtained by raster scanning the image.
3 to convert into a binary image as shown in FIG. 5, 24. The image information of this binary image is written onto the image memory in l) MA mode by an image information input circuit 14 including an image memory and an image feature extraction section. The connectivity analysis unit 15 extracts line segment information (coordinates of the end of the line segment, length, presence or absence of overlap with adjacent line segments) of the binary image in the image memory extracted from the image feature extraction unit. After scanning one screen, the system analyzes the connectivity of line segments and assigns the same pattern number to line segments that belong to the same independent pattern. , separation between independent patterns is performed. The boundary point detection unit 16 detects the boundary point coordinates of the independent pattern together with the pattern number to which the boundary point belongs. This boundary point information is analyzed by the outer/inner circumference point tracking calculation unit 17, and by removing the outer circumference point sequence data from the counterclockwise (or clockwise) outer circumference point sequence data and the boundary point information, the inner circumference point Column data is obtained.

Note that this inner periphery point sequence data is used when many circular objects are superimposed and a part of the outer periphery of the circular object appears on the inner periphery of the overlapping pattern, so if there is no such possibility, this No data extraction or subsequent processing is required.

The subsequent processing for the inner circumference point sequence data is exactly the same as for the outer circumference point sequence data, and hereinafter, when writing outer/inner circumference, it means either the outer circumference, the inner circumference, or both.

Further, the following description will mainly be made regarding the outer periphery.

Next, the radius and center coordinates of each individual arc on the outer/inner circumference are calculated in the arc calculation unit 18 from the outer/inner circumference point sequence data of the superimposed circular image. From these individual arc data, a set of individual arc data belonging to the same circle (same circle/individual arc group data) is created. From this same circle/individual arc group data, the estimation calculation unit 19 calculates the position and size of the estimated circle (also called minimum or object circle). This makes it possible to obtain three pieces of accurate information: the number, position, and size of the circular objects that make up the image. These detection results are output from the detection result output unit 20 to the outside as object detection data 21, and are
It is output to 110 devices such as T.

For circular objects, whether they are non-superimposed objects or objects that are superimposed, as long as they are not completely overlapped and cannot be seen, if a part of the ring part can be grasped by the imaging means 12, this detection device 22 The whole can be estimated by

A detailed explanation of each of the above-mentioned parts will be given below.

FIG. 5 shows a binary image 24 of the superimposed circular object after being binarized via the binarizing circuit 13. A coordinate system X, Y is defined within the effective screen 23, and the image information input circuit 14
The value of the geometric information of the image can be manually stored in the image memory.

FIG. 6 shows an example of a binary image containing independent patterns 25 and 26 of two objects. The "o" marks indicate object points 27 as pixels within the pattern. The independent pattern 25 has no holes, and the independent pattern 26 has one hole 261.

FIG. 7 shows a 3×3 mask M consisting of eight adjacent pixels centered on a point of interest A as a pixel of interest, where B is called a 4-neighboring point and C is called an 8-neighboring point. Note that the object points 27 are classified into the following five types.

0) Interior point: If all the object points 27 are in the 4 neighboring points B, (b) Left point: If there is no object point 27 on the left side, ii) Cloth fish: If there is no object point 27 on the right side, ii) Top Point: Neither the left point nor the fish, and there is no object point 27 on the upper side, (e) Bottom point: Neither the left point nor the fish (, object point 27 on the bottom
If there is no , the left point, the upper point, and the lower point are now boundary points (contour points).

As mentioned above, a series of object points 27 on one horizontal scanning line is called a line segment, and based on the line segment information extracted by the image information input circuit 14, the connectivity analysis unit 15 calculates the number of points on one screen as described above. After scanning is completed, the connectivity before and after each line segment is analyzed, and the pattern attachment force for each independent pattern to which the line segment belongs is given to each twin segment. Next, in the boundary point detection section 16,
It is possible to know which independent pattern each boundary point belongs to, based on the boundary points including the outer circumferential point and the inner circumferential point of each independent pattern. Next, in the outer/inner circumference point tracking calculation unit 17, from the boundary point data of each independent pattern, outer circumference point sequence data and inner circumference point sequence data for each independent pattern, that is, boundary points of the outer or inner circumference (hole 261). The coordinates of , and data in which direction codes, which will be described later, at the coordinates are arranged in order of adjacency are calculated.

In other words, counterclockwise (or clockwise) tracking is performed of the boundary points (left point, upper point, lower point) for each independent pattern. Here, the term "counterclockwise (right) rotation" refers to the direction in which the pattern is tracked while looking at the pattern to the left (right) side, while passing between the outer and inner peripheries of the outline of the pattern.

Figure 8 shows the direction code 1-0. This is the stage of tracking the boundary points, and depending on the relationship between each layer point A and the position of the next adjacent boundary point, the direction code is 1-0 as shown in the figure. The number 8 represents the tracking direction.

Various methods are known for this tracking procedure, and for example, the method described in Japanese Patent Application No. 58-9108 "Pattern Contour Tracking Method" by the present applicant can be used. FIG. 6 shows the results of counterclockwise tracking of the outer/inner circumferential points for each independent pattern 25°26. In this way, for each independent pattern, first, by tracing the contour points on the outer periphery, the outer periphery point sequence data consisting of the coordinates of each outer periphery point and the direction code at the point is obtained, and then for the remaining contour points, as the inner periphery. By tracking, the inner circumferential point sequence data is obtained.

Next, the arc calculating section 18 will be explained. This section includes a convex image separation section, an outer/inner circumference/representative point selection section, and an individual arc calculation section.

First, the operation of the convex image portion #age will be described.

The direction code in the outer circumferential point sequence data is as shown by the solid arrow in Figure 9. In the convex independent pattern 91, the direction code only rotates counterclockwise (increasing direction, which will be described later); Independent pattern 9 with shaped parts 921 and 922
2, the groups 921 and 922 perform a reverse rotation (rotation in the decreasing direction, which will be described later).

In addition, in the case of the direction code of the inner circumference point sequence, pattern 9 in Fig. 9
2 is a hole pattern (therefore, it is assumed that the outer pattern for this pattern is outside the figure), the direction code changes as shown by the dotted arrow, and the concave portion 92 described above changes.
1,922 is replaced with a convex portion in this case, and the remaining convex portions of −Jd are considered to be concave portions. At this time, the direction code rotates counterclockwise (increasing direction) in the convex portions 921 and 922 while tracking the outer periphery, and reverses (rotates in the decreasing direction) in the remaining concave portions.

FIG. 10 shows images of three circular objects 2B, 29, 30. The number of independent patterns is two, 101 and 102, because the circular objects 29 and 30 overlap.

The 1h of outer circumferential points are determined for each independent pattern using the procedure described above. Therefore, in the case of "independent pattern 101", the direction codes are consecutively 1 → 2 → 3 → 4 → 5 → 6 → 7 →
It changes in the direction of 8→1→2→... (increasing direction). The direction code changes from 8 → 7 → 6 → 5 → in the middle of outer circumferential point tracking.
When searching for a place where the change occurs in the direction of 4 → 3 → 2 → 1 → 8 → 7 →... (decreasing direction), we find intersection points 31 and 32 where the boundary points of images of circular objects 29 and 30 intersect. be able to. The relationship ``/work'' applies in exactly the same way to the case where the intersection point of a circular object is attached on the inner periphery.

However, when the contour tracking direction is clockwise, the increasing direction and decreasing direction are interchanged.

In this way, the boundary points near the intersections 31 and 32 are deleted, and the boundary point data consisting only of the data of each convex image part is obtained by the coordinates (xi
, yi), and at the same time, it is regarded as a vector (FIG. 10 Vi) from the coordinate origin O to the coordinate Pi of the boundary point. Note that the boundary points will be called by these coordinates below.

In addition, Ni is the number of the independent turn 102 to which the boundary point Pi belongs, Fi is the direction code from the boundary point Pi to the adjacent boundary point, and i is the sequence of boundary points on the same independent turn (
These numbers are sequentially assigned to the outer circumference point sequence or the inner circumference point sequence.

As a method for detecting a convex image portion on the contour of such a pattern, the above method has the advantage of shortening the calculation time, but another method is described in Japanese Patent Application No. 1983 filed by the present applicant. -76771 “Contour feature detection method”
It is also possible to use a change in sign of the contour curvature at .

Next, in the individual arc calculation section, the data (coordinates, radius) of the center of the individual arc corresponding to the boundary point PK is obtained using the following procedure.
I put it on.

Here, if we look at the layers of the same independent pattern, the boundary point data should be (Pi, F'iJ).
Next, as will be described later, three boundary points Pi-1 and Pi are selected by the outer/inner circumference/representative/side selection section.

Consider Pi+1 (both shown in Figure 10). The line segment between each point is vector σ1. If expressed as σ2, σl = P i - P (i -1') C2 = 1' (i + D - P i.Here, when 1011 = 1σ21, that is, the lengths of the two line segments are equal Then, the radius R of the circular arc (individual circular arc) passing through the three boundary points is determined by the following formula.

kL? = S/ 2- sin (9'/21・-・-
・(11) Here, S is the absolute value of the vector σ1 or C2 (the length of the line segment representing the vector). That is, s=1σ11−1σ21 ・・・・・・・・・・・・・・・
...... (2), and ψ is the vector σ1
(If the central angle of the above-mentioned arc with respect to C2 is 91° ψ2 as shown in Figure 10, then ψ=ψ1−ψ2 ・・・・・・・・・・・・・・・・・・・・・
(3), and these will be discussed later.

Here, 1) IIIf' of the outer/inner circumference/representative point selection section
When we multiply Σ, we get Pi-1 as the three points that enclose the individual arcs.
1Pi, Pi+1 as the outer periphery point sequence (Pi) (outer periphery point IJ
The selection method is such that the total length and absolute value of the &-1 vector σl and C2 are the same or approximately equal. In this case, instead of calculating the outer circumference point sequence (P i '1) for each arc like this, 7', C (l! Select preliminary representative points such as σ), and from among them, determine the non-representative Jfd AL that is a set of three points as described above.In this way, the number of times of Then, kone, raσ) So that the representative peripheral points are approximately equally spaced, the representative points are approximately equidistant using the well-known method of calculating the approximate perimeter using the "Method 111" code. Select as a peripheral point. -4-1 Buwachi σ from the boundary point Pi-1 to the point Pi]
Outside) Moon, @1jII (P i ) (P i ) (line of boundary points), trace to the point where the chord force '-1°3.5.7 is the horizontal map (or vertical) jll' The number of points corresponding to the no direction code force '-2, 4, 6, and 8 is traced in the diagonal direction. When the number σ) is ne, the length of the circumference and the length of the arc from point P i-1 to point Pi are relatively short, so it is a simultaneous trap.

The length of the chord between them, that is, the length s of the vector σ1
l (the above 8 is further divided and named like this), and the above formula (2) is S i = l a 11 = i
P i −P<1-1)l;no +JN ne−=
(2) becomes -1 and the length Sl is determined. Boundary point Pi+
1 is the circumference length from the point Pi, and therefore the length s2 of the vector σ2 (dividing S in the same way as above) is selected so that it has a value (distance) close to this length Sl. The length S2 can be determined in the same manner as in formula -1.

Next, returning to f) and the explanation of the individual arc calculation section, if this is done, the length of the vector in equation (2) above will be the length Sl.

It can be regarded as the average value of S2, and by setting S=(S1+82)/2, the above equation (1) is transformed as shown in the following equation (1)-1 to find the radius R.

! R0= (S1+82)/4 ・5in(ψ/2)...
...(1)-1 Here, for 5 inches (ψ/2), if the relationship in equation (3) above exists, then the angle ψ is the 10th
Since it is equal to the angle formed by the vectors σ1 and C2 in the figure, cosψ is obtained from the equation (4) below by calculating the scalar product (inner product) σl・C2 of the vectors σ1 and C2, and c
osψ=σ1・C2/S1・S2・・・・・・・・・・
...(4) Use the relationship of the following equation (5) in this equation (4).

5in(ψ/2)-J(1-CO5-9+)/2...
−(51 Note that here, the scalar product σJ・C2 of equation (4)
are respectively σ1x+σ2x, σ1y, and σ2y as valley vectors σ1. When the X-axis and Y-axis components of σ2 are, σl x= (x i −xti −1)), 02
y=(xti+1l-xi)C1y=(yi-y(
i −11), “2y=(X(i+1l−yi)
It is calculated by the following formula.

σIIIσ2−σ1xIIσ2x+σ1y・C2y=(
xi −x(i−D)(x++p−xi)+(yi
−y(i−1(yi+1M”) 1On the other hand, the center coordinates C1(xoi) of the individual arcs in FIG.

yoi) (this can also be thought of as a vector from the origin O to the coordinate C1) can be calculated using the following formula.

C1=((C2-a l )/41σ2-σ11)・)
L.
Since it is represented by l, the point P! Is the vector from u to the center Ci of the arc u −R? = ((a2-a1)/Iσ2-σmonths・R?1 This is a relationship using the following equation.

Above, the representative peripheral points (boundary points) Pi −1, Pi ,
The center coordinates Ci and radius R0 of the individual arcs are calculated from Pi+103 points. Here, the central point Pi of the three points mentioned above is called an individual arc representative point.

Note that another method for finding the center of each individual circular arc is the method shown in FIG. This is the line segment d1 connecting the boundary points Pi-1 and Pi.
By using a method in which the center Ci of the individual arc is the intersection of the perpendicular bisector h1 of Let the distance be the radius R1! It is. In this method, the line segments d1 and d2 do not need to be approximately equal. In this way, the boundary point data (
From the representative peripheral point (Fil (Pi, Ni, Fi)
Next, the estimation calculation unit 19 will be explained. This part includes a portion of the same circle/individual arc group and a least squares circle calculation part, and first, when the above-mentioned individual circle t data with different numbers i and j are compared in the separation part, Divide into those that satisfy the following conditions.

1 几, −, a, 1 <ΔR (constant minute value) 1ci−Cjl<ΔC (constant minute value) In other words, it is divided into a set of circular arcs with approximately equal radii and close center coordinates. In this way, the same circle/individual arc group data (Ci, R,, Nc
iJ is obtained. Here, Nci indicates a number for estimation (least squares circle, object circle) to which the individual arc with number i determined in the above procedure is considered to belong.

Next, the least squares circle calculation unit calculates the estimated inner push data that is considered to belong to the same estimation, ((-+, i(·, , NCI.

Using Pi, Pi-1, Pit, and Nil (however, each internal code is as described above), the estimation coordinate Co and radius Ro are estimated 1- by the least squares method.

In other words, in Fig. 12, the representative outer circumferential point (coordinates) is P1
〜P6, and the center coordinates and radius of the arc formed from the three points (PL, P2.P3) are respectively (C2JL2), and similarly the points (P2.P3.P4), (P3.P4.P5)
, (P4.i'5.P6), the center coordinates of the arc,
When the radii are (C3, 几) and (C4, 几, ), respectively, the center CO2 radius 几0 for estimation is calculated from the center coordinates C2 to C4 of the arc, the radius R2 to R4, and the coordinates P1 to P6 of the outer circumferential point.
It is something to be used for.

Therefore, the center CO for estimation and each individual arc representative point P+
A method (least squares method) is used to find the estimated radius and center coordinate CO that can minimize the sum of the squares of the residuals (errors) obtained by subtracting the estimated radius O from the distance Ri. That is, 2
The sum of multiplicative errors is expressed by the following formula.

1 = X (Ri − Ro)・・・・・・・・・
・・・・・・・・・・・・・・・・・・(7) Therefore, the sum■
Minimize the estimation radius 1-LO, center coordinate CO (
-=(xoryo)), it is necessary to satisfy the following formula.

δ1/δxg=δ■/δyO=bI/δRO= 0 ・
...-+81 Therefore, the following three forces can be obtained from equation 8).

J((Ri−几o)/Ri)(xi−xo)=O・・−
(911Σ((R1-Ro)/Ri)(yi-yo)
=0・・・・・・(9)−25ha(Rule-Ro)/Ri)=O・・・・・・・・・・・・
+9)-3 Solve the above three equations (9)-1, (9)-2, and (9)-3~・Center coordinates (xo, yO), radius 9003
Enter the value of ('f.

good. However, the above equation becomes a transcendental equation that includes an irrational number, and is simply j! #Kena(・So, from these three equations, we derive the following successive equations using the successive approximation method.

In other words, when the center coordinates and radius for estimation as a solution found in the first approximation are respectively expressed as (XS + '/'0) +P, the estimate n+1 n+ which is the solution of the (n+1)th approximation is
1 n+1 Center coordinates of fixed day, radius (xotY LR0&t, x”'
=(J xi +R", -x", -X(1/1(7)
-R”,・J(xi/1(7))/ No 1 1 ・・・・・・・・・(10) −1 ・・・・・・・・・(10) −Madashi here P is the center coordinate for the first estimation ('+
y") and the distance between each of the individual arc representative points Pi, 0
0 In other words, it↑= (xi-x")"+ (exploitation7y...
...(10) -4100 where N is the number of the individual arcs.

Next, when calculating the above sequential equations (10)-1 to (10)-4, in order to speed up the convergence of the solution, the first-order solution is calculated for each of the individual circular arcs as shown in the following equation. radius R.

And its center coordinate C! Proceed with the calculation as the average value. That is, the center coordinates (xo) of the first estimated circle.

y), radius R is 0 x = (guest xOi)/N ・−・−・・・−(11)
-1o.

Y=C:yOi]/N ・・・・・・・・・(11)
-20 鳳几＝〔GI几）/N ・・・・・・・・・(11) -3
11, the calculation proceeds in the order of 2nd, 3rd, . . . nth. In this way, the value of n can normally be converged in 1 to 5 times.

In this way, the size and position of the estimated circle (least squares aperture) can be determined with high accuracy. From the above results, the number, size, and position of circles included in the image are known, and this is sent to the detection result output unit 2.
0 and is output as the determination result in-object detection data 21.

〔Effect of the invention〕

First, to summarize the functions of the main components of the present invention as described above, including their effects: ■ The connectivity analysis unit 15 separates the binary image into independent turns and processes each of these patterns. For independent circular objects, the circle can be estimated correctly.

■ The boundary point detection unit 16 and the outer/inner circumference point tracking calculation unit 17 track the outline of the binary image of the superimposed circular object, and
Outer/inner point sequence data organized as an inner point sequence (Pi
, Fi , Ni ), the data can be separated into outer/inner point sequence data for each convex image part, and the non-representative/inner points are selected from each data and the corresponding Individual arc group data of center coordinates and radius of each individual arc (
Ci.

R,) is obtained.

■ In the estimated circle calculation unit, the individual circular arc group data is separated into sets of center coordinates and sizes of each individual arc, and for the same circle/individual circular arc group data obtained in this way, a minimum of 2 Calculate the estimated circle as a multiplicative circle.

In this way, by focusing on the contour points of the image, it is possible to classify them in multiple stages from a point set to a circular arc set, a circle set, etc., and quickly estimate the inside of the object.

In this way, according to the present invention, the following positions, sizes, and numbers of circular objects can be known.

(b) It is possible to know the position and size of any object with a circular outline, excluding holes.

(b) Even if circular objects overlap, part of the outline
If it appears in a binary image, it can be detected separately.

If it is desired to know the position of the circular arc portion included in the C1 external shape, it can be detected.

) It is possible to exclude objects that are not circular.

) By estimation using the method of least squares, even if the shape is not a smooth circular shape, that is, even if the edge is jagged and the circle is used as an envelope, a circle that approximates the shape, therefore, becomes the envelope. Yen can also be estimated.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing the configuration of a conventional circular object detection device, FIGS. 2 and 3 are diagrams showing examples of images of the device shown in FIG.
FIG. 4 is a block diagram showing the configuration of the present invention, FIG. 5 is a diagram showing a binary image when circular objects are superimposed, FIG. 6 is a diagram showing boundary points of a general binary image, and FIG. is a diagram explaining the definition of connectivity using a 3 × 3 mask, Fig. 8 is a diagram showing the direction code of the outer peripheral point, and Fig. 9 is a diagram showing the direction code of the contour point counterclockwise when the direction code moves. Figure 10 is a diagram explaining that there is no reversal in the decreasing direction.
Figure 1 is a diagram showing another method of determining the center and size of an individual arc, and Figures V and 12 are diagrams explaining a method of determining an estimated circle (least squares circle). Description of symbols 11... Circular object, 12... Imaging means, 13... Binarization circuit, 14... Image information input circuit, 15... ... Connectivity Analysis Department, 16.
... Boundary point detection section, 17 ... Outer/inner circumference point tracking calculation section, 18 ... Arc calculation section, 19 ...
...Estimated circle calculation unit, 20...Detection result output unit, 21...Object circle detection data, 22...
・Circular object detection device. T Fang 6 Fig. Fang 7 Fig. Fang 9 Fig. Fang 12 Fig.

Claims (1)

[Claims] 1) After raster scanning images of a plurality of objects whose contours have circular arcs or circular arcs as envelopes, 2)
In an apparatus for detecting the position and size of the circular object by converting it into a binarized image through a digitizing means, the coordinates of the end of the scanning segment of the pattern of the circular object in the binarized image, an image feature extracting means for detecting an overlap with a previous scanning segment, and adding a pattern number for each independent pattern to all scanning segments of the binarized image based on the pattern information by the image feature extracting means. The connectivity analysis means traces the boundary points one time in the front counterclockwise (clockwise) direction using the coordinates of the end of the scanning segment having the same pattern number, and calculates the coordinates of each boundary point arranged in the tracking order, Direction code indicating the tracking direction from a boundary point to an adjacent boundary point Outer/inner point tracking calculation means that tracks along the inner circumference in the same manner as the outer circumference and outputs either or both of the outer point string data and the same inner point string data; a convex image section for extracting loci-like divisions of the outer/inner circumference point sequence of the convex image section in tracking the boundary points from the outer/inner circumference point sequence data and outputting the resultant data as outer/inner circumference/convex point sequence data; The separating means and the outer/inner points constituting the outer/inner periphery/convex point sequence data are sequentially separated into a plurality of individual arcs passing through three successively adjacent points at adjacent outer/inner periphery representative points. Individual arc calculation means for calculating and outputting each center coordinate and radius; and same circle/individual arc calculation means for separating those individual arcs whose center coordinates and radius values are close to each other as belonging to the same circle. , estimated circle calculation means for calculating and outputting estimated center coordinates and radius using each data corresponding to each of the individual arcs belonging to the same circle. 2. In the detection device according to claim 1,
The estimated circle calculation means uses the average values of the center coordinates and radius of each individual arc as first approximations of the center coordinates and radius of the estimated circle, and calculates the outer/inner representative points corresponding to each individual arc. Perform successive approximation of the required order to minimize the sum of the squares of the residuals obtained by subtracting the radius of the estimated circle from the distance between the coordinates of a point consisting of at least a portion of the estimated circle and the center coordinates of the estimated circle. Characteristic circular object detection device. 3) 4? In the detection device according to claim 1 or 2, the convex image portion separation means is configured to detect a value of the direction code in the outer/inner circumferential point sequence data that increases (decreases) in the order of the tracking. ) A circular object detection device characterized in that it is a means for extracting coordinates of outer/inner circumferential points in a section.