JP5027783B2 - Circle detector - Google Patents

Description

  The present invention relates to a circle detection device that detects a circle from an input image.

Conventionally, as a circle detection device that detects a circle from an input image, for example, there is a circle detection device disclosed in Patent Document 1. The circle detection apparatus first performs a process of removing a linear component from an input line drawing, and then obtains an intersection of the input line drawing from which the linear component has been removed, a scanning line in the x direction, and a scanning line in the y direction. The circle to be searched is extracted from the figure drawn by the midpoint between them.
Japanese Patent No. 3694911

  In the circle detection device disclosed in Patent Document 1, if there is a break in the circular figure in the input line drawing, the midpoint cannot be detected correctly, and the circular figure may not be extracted correctly.

  As a method for extracting a figure from a discontinuous input line drawing, an extraction method using Hough transform is conventionally known. In the Hough transform process, a point is extracted from each point on the figure in the input line drawing to a parameter space consisting of parameters representing the figure, and the maximum value is obtained in the parameter space to determine the parameter and extract the figure. It is. When a circle is detected using such a Hough transform process, in order to obtain three parameters of the circle center position (a, b) and radius r, voting to the abr parameter space (three-dimensional space) is performed. The above three parameters are determined by determining the maximum value, but there is a problem that the processing time and the memory used increase because the processing of voting to the three-dimensional space and the search for the maximum value is necessary. .

  The present invention has been made in view of the above problems, and an object of the present invention is to provide a circle detection apparatus that can perform a process of detecting a circle from an input image at a high speed and can use a small amount of memory. There is to do.

  In order to achieve the above object, the invention according to claim 1 is characterized in that an edge image extraction means for performing an differentiation process on an input image and extracting an edge point having a differential intensity greater than a predetermined threshold value is extracted. Voting means for performing voting only on θ in a predetermined angle range centering on the edge direction at the edge point when performing Hough transform processing for straight line detection on the edge point and performing voting processing on the parameter space (ρ, θ) Voting cell extracting means for extracting cells voted from the (ρ, θ) parameter space, and two sinusoidal substantially parallel curves continuous in the θ direction from the cell group extracted by the voting cell extracting means. And a center for obtaining the center position of the search target circle from the shape of the intermediate line equidistant from the two curves. Position detection Characterized by comprising a stage. Here, the edge direction is a direction in which the density of the edge changes, and is a direction orthogonal to the tangent of the circle at the edge point.

  According to a second aspect of the invention, in the first aspect of the invention, the voting cell extracting means extracts a cell whose voting value in the (ρ, θ) parameter space is within a predetermined range. Circle detection device.

  According to a third aspect of the present invention, in the invention according to any one of the first and second aspects, the diameter detecting means uses the ρ direction interval between cells at the same θ as the class for the cell group extracted by the voting cell extracting means. A histogram is created, and the ρ-direction interval at which the frequency is maximal and equal to or greater than a predetermined threshold is obtained as a diameter.

  According to a fourth aspect of the present invention, in any one of the first to third aspects, the center position detecting means uses the ρ value of the intermediate line when θ is 0 degrees as the x coordinate of the center position, and θ is 90 degrees. In this case, the ρ value of the intermediate line is the y coordinate of the center position.

  According to a fifth aspect of the present invention, in any one of the first to third aspects of the present invention, the center position detecting means converts the coordinates (ρ, θ) of each point on the intermediate line to a straight line passing through the point (a, b). Substituting into the equation a · cos θ + b · sin θ = ρ, the values of a and b satisfying the straight line equation are obtained as the coordinates (a, b) of the center position.

  According to the first aspect of the present invention, the voting means enters the (ρ, θ) parameter space only for θ in a predetermined angle range centering on the edge direction at the edge point, that is, the direction orthogonal to the tangent of the circle at the edge point. Voting is performed, and the diameter detection means obtains two sinusoidal substantially parallel curves continuous in the θ direction from the cell group extracted by the voting cell extraction means. The ρ value at the angle θ of these two curves is the distance between the two tangents and the origin such that the angle of a perpendicular (ie, a straight line parallel to the edge direction) perpendicular to the tangent of the circle is θ, In addition, since the absolute value of the difference between the ρ values of the two curves at the angle θ is the diameter of the circle, the diameter detecting means can determine the diameter of the circle from the interval between the two curves in the ρ direction. Further, since the intermediate line equidistant from the two curves becomes a figure in which all straight lines passing through the center position of the circle are mapped to the (ρ, θ) parameter space, the center position detecting means uses the shape of the intermediate line. Based on this, the center position of the circle can be obtained. Therefore, without voting to the three-dimensional parameter space as in the conventional circle detection method, voting to the (ρ, θ) parameter space only for θ in a predetermined angle range centered on the edge direction for each edge point. This makes it possible to detect a circle, so that the circle search process can be performed at a high speed and the amount of memory used can be reduced. In addition, since the diameter and center position of the circle are obtained from the two curves obtained by performing the Hough transform processing for line detection, the circle is accurately detected even if the circle to be searched is interrupted in the input line drawing. There is an effect that can be.

  According to the invention of claim 2, since the voting cell extracting means extracts cells whose voting value is within a predetermined range, the voting cell extracting means performs processing for detecting the diameter and center position of the circle by removing unnecessary cells. It can be performed at high speed and circles can be detected efficiently. Further, in the circle detection device of Patent Document 1 described above, the process of removing the linear component from the input line drawing is performed. However, according to the invention of claim 2, by extracting the cells whose voting value is within the predetermined range, Since unnecessary cells are removed, processing for removing a straight line component from the input line drawing is not required, and a circle search can be performed at higher speed.

  According to the invention of claim 3, the diameter detecting means obtains the diameter of the circle based on the result of creating a histogram based on the result of creating a histogram with the ρ direction interval between cells at the same θ as the class. Even if there is a variation in the interval, or there is a break in the search target circle or the extracted curve, the diameter of the circle can be accurately detected.

  According to the fourth aspect of the present invention, the center position of the circle to be searched can be obtained only by obtaining two points on the intermediate line, so that the center position can be detected at high speed.

  According to the invention of claim 5, by substituting the coordinates of many points on the intermediate line for the straight line expression passing through the point (a, b), the values of a and b satisfying the straight line expression, that is, the center position Since the coordinates are obtained, the center position of the circle can be obtained with high accuracy even if the search target circle or the extracted curve is interrupted.

  Embodiments of the present invention will be described below with reference to the drawings.

  The circle detection apparatus A of the present embodiment is used for detecting a circular shape from an image captured by the imaging means 1 such as a CCD camera, for example, and as shown in the block diagram of FIG. An image storage means 2 for storing image data input from the means 1 in an image memory (not shown), and an input image from the image pickup means 1 stored in the image memory is subjected to a differentiation process so that the differential intensity is predetermined. Edge image extraction means 3 for extracting edge points larger than the threshold value of the image, and Hough transform processing for line detection is performed on the extracted edge points to perform voting processing on the (ρ, θ) parameter space In addition, voting means 4 for voting only in a predetermined angle range θ centered on the edge direction at each edge point, voting cell extracting means 5 for extracting a voted cell from the (ρ, θ) parameter space, and voting cell Diameter detection means for obtaining two sinusoidal substantially parallel curves continuous in the θ direction from the cell group extracted by the extraction means 5 and obtaining the diameter of the circle to be searched from the interval in the ρ direction of the two curves. 6 and center position detecting means 7 for obtaining the center position of the circle to be searched from the shape of the intermediate line equidistant from the two curves.

  A circle detection process by the circle detection apparatus A will be described with reference to FIGS. FIG. 2A shows a monochrome grayscale image F1 captured by the imaging means 1, and the image data of the monochrome grayscale image F1 input from the imaging means 1 is stored in the image memory by the image storage means 2.

  The edge image extraction means 3 reads the image data of the black and white grayscale image F1 from the image memory, and executes differential processing using, for example, the Sobel filter shown in FIGS. 3A and 3B for each pixel of the black and white grayscale image F1. Then, by extracting an edge point whose obtained differential intensity is equal to or greater than a predetermined threshold value, an edge image F2 as shown in FIG. 2B is created. 3A is a filter for extracting a vertical component (horizontal differentiation), and FIG. 3B is a filter for extracting a horizontal component (vertical differentiation). Then, the differential intensities in the horizontal direction and the vertical direction are obtained using two filters, and the edge direction of each pixel is calculated from the differential intensities in the horizontal and vertical directions. Note that a circle C and a straight line B appear as graphic elements in the edge image F2 shown in FIG. 2B, and the circle C that is a search target is executed by executing the following processing in the circle detection device A of the present embodiment. The center position and diameter of the are detected.

  The voting means 4 performs a Hough transform process for straight line detection on each edge point on the edge image F2 shown in FIG. 2B, and performs a voting process in the (ρ, θ) parameter space. Voting is performed only on θ within a predetermined angular range (ω−Tw <θ <ω + Tw) centering on the edge direction ω at the edge point. That is, when the coordinates of the edge point are (p, q) and the edge direction is ω, voting is performed on a cell corresponding to (ρ, θ) obtained by the following equation (1).

ρ = p × cos θ + q × sin θ (where ω−Tw <θ <ω + Tw) (1)
Here, Tw is an allowable angle value with respect to the edge direction ω, and voting is performed only on θ in the angle range larger than (ω−Tw) and smaller than (ω + Tw) with the edge direction ω as the center. Note that the edge direction ω is a direction in which the density of an edge changes, and in the case of an edge point on a circle, the direction is perpendicular to the tangent of the circle at the edge point. Also, the smaller the radius of the circle to be searched is, the larger the deflection angle per pixel is, so it is necessary to increase the angle tolerance Tw. For example, if the radius of the circle is about 80 pixels, the angle tolerance It is preferable to vote only for θ within an angular range of (ω−5) <θ <(ω + 5), with Tw being 5 degrees.

  FIG. 4B shows a Hough voting space ((ρ, θ) parameter space) after the voting means 4 performs the above voting process, and the voting process described above for each edge point on the circle C. As a result, two sinusoidal substantially parallel curves D1 and D2 continuous in the θ direction are created, and a cell group E is created by voting the edge points on the straight line B. In FIG. 4 (b), the curves D1 and D2 and the cell group E are shown to be uniform in density for the sake of simplicity, but actually the density varies depending on the number of votes. The number of votes increases as the pixel value increases (darker).

  Here, the voting means 4 performs a voting process to the (ρ, θ) parameter space for all edge points on the edge image F2. Among the edge points on the circle C, the edge direction ω is θ1. A case where the voting process is performed on the edge points P11 and P12 will be described with reference to FIGS. Of the two edge points P11 and P12 whose edge direction is θ1, the one closer to the origin is defined as P11.

  At the point P11, when the voting means 4 votes for θ in a predetermined angle range centered on the edge direction θ1, the tangent line L11 passing through the edge point P11 at the edge point P11 and the tangent line L11 at the edge point P11 are converted into the angular range. A straight line inclined within (θ1 ± Tw) is mapped to the (ρ, θ) parameter space, and the ρ value when mapping the tangent L11 is the distance ρ11 from the origin to the tangent L11. At the point P12, when the voting means 4 votes for θ within a predetermined angle range centered on the edge direction θ1, the tangent line L12 passing through the edge point P12 at the edge point P12 and the tangent line L12 at the edge point P12 are A straight line inclined within the range (θ1 ± Tw) is mapped to the (ρ, θ) parameter space, and the ρ value when the tangent L12 is mapped is the distance ρ12 from the origin to the tangent L12. Here, since the absolute value | ρ12−ρ11 | of the difference between the distance ρ11 to the tangent line L11 and the distance ρ12 to the tangent line L12 is equal to the diameter of the circle C, the ρ value is an intermediate value ρ10 between the ρ11 and ρ12 at the angle θ1. The straight line L10 that becomes (= (ρ11 + ρ12) / 2) passes through the center position P0 of the circle C.

  Therefore, when the voting means 4 performs the above voting process on all the edge points on the circumference, as shown in FIG. 4B, two sinusoidal substantially parallel lines continuous in the θ direction. The curves D1 and D2 are created in the (ρ, θ) parameter space, and the sinusoidal intermediate line D0 (see FIG. 5C) that is equidistant from the two curves D1 and D2 is represented by the circle C All straight lines passing through the center position P0 are mapped to the (ρ, θ) parameter space.

  Next, when the voting process by the voting means 4 is completed, the voting cell extracting means 5 has extracted a cell group having a voting value, that is, a cell group having a voting value of 1 or more, as shown in FIG. Three cell groups G1, G2, G3 are extracted. Here, the cell groups G1 and G2 are cell groups obtained by mapping the edge points on the circle C to the (ρ, θ) parameter space, and the cell group G3 is the edge point on the straight line B with the (ρ, θ) parameter. It is a group of cells created by mapping to space.

  As shown in FIG. 5 (b), the diameter detecting means 6 applies the same labels LB1, LB2, LB3 to the cells adjacent to the cell groups G1, G2, G3 extracted by the voting cell extracting means 5, respectively. After the application, the difference between the maximum value and the minimum value of θ is obtained for each cell of each label, and the labels (in the example shown, labels LB1, LB2) where the difference between the maximum value and the minimum value is a predetermined threshold (for example, 150 °) or more. ) Is extracted, only the cell groups G1 and G2 formed by mapping the edge points on the circle C are extracted. Here, since the edge directions of the edge points on the straight line B are substantially the same direction, the cell group formed by mapping the edge points on the straight line to the (ρ, θ) parameter space has a range of angles θ of a circle. It is considered that the cell group formed by mapping a straight line can be removed by comparing the difference between the maximum value and the minimum value of θ with the threshold value.

  When the label extraction is completed, the diameter detection means 6 applies a straight line to each of the extracted labels LB1 and LB2 by the least square method, and the labels whose inclination difference is within a predetermined threshold are substantially parallel curves. In the example of FIG. 5B, the cell groups G1 and G2 of the labels LB1 and LB2 are extracted as two substantially parallel curves D1 and D2 (see FIG. 5C). When the cell group corresponding to the curves D1 and D2 is extracted by the above processing, the diameter detecting means 6 uses the absolute value of the difference between the ρ coordinates in the same θ coordinate in the two curves D1 and D2 as the diameter. Looking for. In the diameter detection means 6, the absolute value of the difference between the ρ coordinates of the plurality of θ coordinates in the two curves D1 and D2 may be obtained, and the diameter of the circle may be obtained by taking the average of these values. Can be reduced.

  When the diameter of the circle is detected as described above, the center position detection means 7 obtains an intermediate line D0 that is equidistant from the two curves D1, D2, and the center position of the circle C is determined from the shape of the intermediate line D0. Ask for. That is, the center position detecting means 7 obtains the ρ value of the intermediate line D0 by averaging the ρ values of the curves D1 and D2 extracted by the diameter detecting means 6 for each angle θ, and on the intermediate line D0. The coordinate value (ρ0, θ0) of the point Cmax where the ρ value is maximum and the coordinate value (ρ1, θ1) of the point Cmin where the ρ value is minimum are obtained. Here, since the intermediate line D0 is a sine wave formed by mapping all straight lines passing through the center P0 of the circle, the center position detecting means 7 associates the coordinates of the points Cmax and Cmin with the center position of the circle in advance. The attached center calculation table (see Table 1) is created and stored in a memory (not shown), and the coordinates of the point Cmax (ρ0, θ0) and the coordinates of the point Cmin (ρ1, θ1) obtained by the above processing are searched. The coordinates of the center P0 of the circle are obtained by referring to the center calculation table as a key.

  As described above, in the circle detection device A of the present embodiment, the voting means 4 has a predetermined angle range θ around the edge direction at the edge point in the edge image, that is, the direction perpendicular to the tangent of the circle at the edge point. Only in the (ρ, θ) parameter space, and the diameter detection means 6 from the cell group extracted by the voting cell extraction means 5, two sinusoidal substantially parallel curves D1 continuous in the θ direction. , D2. Here, the ρ values of the two curves D1 and D2 at the angle θ are two tangents such that the angle of a perpendicular (that is, a straight line parallel to the edge direction) perpendicular to the tangent of the circle at a certain edge point is θ. Since the absolute value of the difference between the ρ values of the curves D1 and D2 at the angle θ is the diameter of the circle, the diameter detection unit 6 uses the interval between the two curves D1 and D2 in the ρ direction. From this, the diameter of the circle can be determined. Further, the intermediate line D0 that is equidistant from the two curves D1 and D2 is a figure in which all straight lines passing through the center P0 of the circle C are mapped to the (ρ, θ) parameter space. Then, the center position P0 of the circle C can be obtained based on the shape of the intermediate line D0. Therefore, without voting to the three-dimensional parameter space as in the conventional circle detection method, voting to the (ρ, θ) parameter space only for θ in a predetermined angle range centered on the edge direction for each edge point. This makes it possible to detect a circle, so that the circle search process can be performed at a high speed and the amount of memory used can be reduced. In addition, for each edge point, voting is performed in the (ρ, θ) parameter space only for θ in a predetermined angle range centered on the edge direction. Since the edge direction is a direction orthogonal to the tangent, the edge direction is the center. By voting to the (ρ, θ) parameter space only for θ in the predetermined angle range, the voting process to the (ρ, θ) parameter space is performed limited to the angle range where there is a high possibility that a tangent exists. Thus, the circle search process can be performed at a higher speed. Furthermore, since the diameter and center position of the circle are obtained from two substantially parallel curves D1 and D2 obtained by performing the Hough transform processing for line detection, the search target circle is interrupted in the input line drawing, Even if the two extracted curves are interrupted, the search target circle can be accurately detected.

By the way, in the voting cell extracting means 5 described above, all cells having a voting value, that is, cells having a voting value of 1 or more are extracted from the (ρ, θ) parameter space after voting by the voting means 4. It is also preferable to extract a cell having a value within a predetermined range. When the edge point on the straight line B is mapped to the (ρ, θ) parameter space in the edge image F2 shown in FIG.
Since the edge directions of the edge points on the straight line B are substantially the same angle, a large vote value is obtained in the (ρ, θ) parameter space (for example, 50 or more), but the circle C is changed to the (ρ, θ) parameter space. When mapping, since the number of edge points where the edge directions are substantially the same is smaller than that of a straight line, the vote value is also small (for example, 20 or less). Therefore, when the voting cell extraction unit 5 extracts cells from the (ρ, θ) parameter space, by extracting cells whose voting values are within a predetermined range (for example, 1 or more and 20 or less), FIG. As shown, only the cell groups G1 and G2 voted by mapping the edge points on the circle C can be extracted, and only the voted cells can be extracted by mapping the edge points on the circle. Can be detected at high speed, and a circle can be detected efficiently. In addition, unlike the circle detection apparatus disclosed in Patent Document 1 described in the related art, it is not necessary to perform a process of removing a linear component from an input line drawing, so that a circle search process can be performed at high speed.

  Further, in the diameter detecting means 6 described above, after obtaining two substantially parallel curves D1 and D2 having a sine wave shape continuous in the θ direction from the cell group extracted by the voting cell extracting means 5, the curve at the same θ coordinate is obtained. The ρ values of D1 and D2 are obtained, and the diameter of the circle is obtained from the absolute value of the difference. However, the diameter calculation method is not limited to the above method, and the diameter is obtained by the method described below. May be. That is, as shown in FIG. 7A, the diameter detection means 6 extracts all cells existing on a certain θ coordinate (θ = α) from the cell group extracted by the voting cell extraction means 5 (θ = α). In the case of α, cells C0, C1, and C2), and among the plurality of extracted cells, processing for obtaining the absolute value D of the difference between the ρ coordinates of two cells (that is, the ρ direction interval between cells) in all combinations Then, the angle α is changed from 0 ° to 180 ° to create a histogram with the value D as a class (see FIG. 7B). Then, the diameter detecting means 6 obtains the ρ direction interval d0 at which the frequency is maximal and equal to or greater than the predetermined threshold Th1 as the diameter based on the result of the histogram, and the ρ direction interval between the curves D1 and D2. Even if there is a variation in the curve D1 and the curves D1 and D2 are interrupted, the diameter of the circle can be accurately detected.

  Further, in the above-described center position detecting means 7, the coordinate value (ρ0, θ0) of the point Cmax where the ρ value is maximum on the intermediate line D0 and the coordinate value (ρ1, θ1) of the point Cmin where the ρ value is minimum. The coordinates of the center P0 are obtained from the center calculation table, but the center position detection method is not limited to the above method, and the coordinates of the center P0 are obtained by the method described below. Also good. That is, in the center position detecting means 7, after the diameter detecting means 6 obtains the intermediate line D0 by averaging the ρ values of the two substantially parallel curves D1, D2, the intermediate line D0 when the angle θ is 0 ° is obtained. The ρ coordinate value ρ (0 °) of the center P0 may be obtained as the x coordinate of the center P0, and the ρ coordinate value ρ (90 °) of the intermediate line D0 when the angle θ is 90 ° may be obtained as the y coordinate of the center P0. Since the center position of the circle to be searched can be determined simply by obtaining two points on D0, the center position detection process can be performed at high speed.

  Further, in the center position detecting means 7, after obtaining the intermediate line D0 as described above, the coordinates (ρ, θ) of all the cells on the intermediate line D0 are obtained, and a straight line expression passing through the point (a, b). By substituting the coordinates (ρ, θ) of all cells into (2) and obtaining the values of a and b satisfying equation (2) by the method of least squares, the coordinates (a, b) of the center P0 are obtained. Also good.

a · cos θ + b · sin θ = ρ (2)
As described above, the center position detection means 7 obtains the coordinates of the center P0 by substituting the coordinates of many points on the intermediate line D0 into the equation (2) of the straight line passing through the points (a, b). Even if the curves D1 and D2 are interrupted, the center position can be obtained with high accuracy.

It is a block diagram which shows the structure of the circle | round | yen detection apparatus of this embodiment. (A) is explanatory drawing of an input image, (b) is explanatory drawing of an edge image. (A) and (b) are explanatory drawings of a Sobel filter. (A) is explanatory drawing of the voting process by a voting means, (b) is a figure which shows the result mapped to the ((rho), (theta)) parameter space. (A)-(c) is a figure explaining the detection process of the circle | round | yen by a circle | round | yen detection apparatus. It is a figure explaining other detection processing same as the above. (A) is a figure explaining another detection process same as the above, (b) is explanatory drawing of the produced histogram. It is a figure explaining another detection process same as the above.

Explanation of symbols

A circle detection device 1 imaging means 2 image storage means 3 edge image extraction means 4 voting means 5 voting cell extraction means 6 diameter detection means 7 center position detection means

Claims (5)

Edge image extraction means for performing differential processing on the input image and extracting edge points whose differential intensity is greater than a predetermined threshold;
When the extracted edge point is subjected to Hough transform processing for line detection and voting processing is performed in the (ρ, θ) parameter space, voting is performed only on θ within a predetermined angle range centering on the edge direction at the edge point. Voting means to perform,
Voting cell extracting means for extracting cells voted from the (ρ, θ) parameter space;
Diameter detection that obtains two sinusoidal substantially parallel curves continuous in the θ direction from the cell group extracted by the voting cell extracting means, and obtains the diameter of the circle to be searched from the interval in the ρ direction of the two curves. Means,
A circle detecting device comprising: a center position detecting means for obtaining a center position of a circle to be searched from a shape of an intermediate line equidistant from the two curves.   2. The circle detecting device according to claim 1, wherein the voting cell extracting unit extracts a cell having a voting value in the (ρ, θ) parameter space within a predetermined range.   The diameter detection means creates a histogram with the ρ direction interval between cells at the same θ as a class for the cell group extracted by the voting cell extraction means, and has a maximum frequency and a predetermined threshold value. The circle detection apparatus according to claim 1, wherein the ρ direction interval as described above is obtained as a diameter.   The center position detecting means sets the ρ value of the intermediate line when θ is 0 degrees as the x coordinate of the center position, and the ρ value of the intermediate line when θ is 90 degrees as the y coordinate of the center position. The circle detection device according to claim 1, wherein:   The center position detecting means substitutes the coordinates (ρ, θ) of each point on the intermediate line into an equation of a straight line passing through the point (a, b), a · cos θ + b · sin θ = ρ, and The circle detection device according to any one of claims 1 to 3, wherein the values of a and b to be satisfied are obtained as coordinates (a, b) of the center position.

Images