JPH0624018B2 - Ellipse detector - Google Patents

Description

TECHNICAL FIELD OF THE INVENTION The present invention relates to an ellipse detection device capable of effectively detecting an ellipse portion in an outline of an object included in a grayscale image.

[Technical background of the invention and its problems]

TV as a part of robot manipulation technology
There is a technique for recognizing an object from a grayscale image captured and input by a camera. For example, when it is difficult for a worker to directly intervene or maintain an object in a dangerous work environment, an image of the work environment is obtained by a remotely controlled TV camera, and this is recognized and processed. After the position of the object is specified, work on the object is performed by remote control.

By the way, in such an object recognition process, for example, when a pipe or the like is viewed obliquely and its connection part (flange) or the like is recognized, it is necessary to detect the elliptical component from the outline of the object. Becomes That is, when a cylindrical body, a columnar body, a disc body, or the like is viewed obliquely, the outer shape thereof is imaged and input as an ellipse, so that it is necessary to detect the ellipse component of the object in the input image.

Therefore, as a method of detecting an ellipse from the outline of an object, it is considered to apply Hough transform of a line segment tracking image. However, if the contour line is a closed curve, line segment tracking can be performed relatively easily, but in reality, the contour line of the object detected in the grayscale image input by imaging is cut off. In most cases. Therefore, it is very difficult to detect an ellipse by tracing a line segment.

The Hough transform is often tried to detect a straight line or a circle, but even when a circle is detected, there are three parameters and the amount of calculation processing is enormous. For this reason, it is barely used when the radius of the circle is known, and is rarely used in practice.

Under such circumstances, in the case of the ordinary Hough transform, it cannot be denied that the elliptic curve having five parameters will further increase the amount of calculation processing, and in practice, there is a problem in its application. is there.

There is also a Hough transform in which the number of parameters is reduced by utilizing the property of an ellipse, but even in this case, it is said that an ellipse component must be present at the target position with respect to the center of the ellipse. There are restrictions. Further, when the line segment forming the ellipse is cut off, the peak is not obtained because the number of data is small, and there is a problem that the ellipse cannot be detected well.

[Object of the Invention]

The present invention has been made in consideration of such circumstances, and an object thereof is to provide an ellipse that can reliably and highly efficiently detect an ellipse forming an outline of an object in a grayscale image. It is to provide a detection device.

[Outline of Invention]

The present invention considers a set of arbitrary two arcs l1 and l2 which do not overlap each other among a plurality of arcs formed by appropriately dividing an ellipse as shown in FIG. , M2 do not intersect with each other, a straight line n connecting arbitrary points of the strings m1 and m2 is a circular arc l.
It is based on the fact that it does not intersect with 1,12.

Therefore, as shown in FIG. 2, there is a possibility that the contour line of the object is extracted from the grayscale image A, the contour line is segmented, and then an ellipse is formed in the segments l1, l2, to ln. A set of segments is extracted as shown in (b) or (e) of the same figure, and its elliptical models p1 and p2 are estimated.

Then, based on the ellipse models p1 and p2, for example, as shown in (c) and (f) of FIG. ) And (g), the elliptic masks q1, q2 have the segments l1, l2, ...
It is verified whether each n is included. Then, the probability of each of the elliptic models p1 and p2 is determined according to the verification result, and the ellipse indicating the outline of the object in the grayscale image is detected with high accuracy and efficiency.

〔The invention's effect〕

Thus, according to the present invention, it is possible to efficiently detect an ellipse from the component of the outline of the cutout of the object detected in the grayscale image. Moreover, the process is advanced while selecting only a set of segments that may form an ellipse from the plurality of segments detected in the image, so that the ellipse detection process can be speeded up.

Also, instead of estimating the elliptic model for all combinations of segments, the whole is predicted from the combinations of two segments and the consistency with the rest of the segments is checked, so that the processing efficiency can be improved. . Furthermore,
Since the accuracy of the estimated elliptic model is verified and the incorrect elliptic model is removed, and only the highly reliable elliptic model is detected, there is a great practical effect such as the improvement of the detection accuracy. Played.

[Embodiment of the Invention] An embodiment of the present invention will be described below with reference to the drawings.

FIG. 3 is a schematic configuration diagram of the embodiment apparatus. The grayscale image captured and input through the image input device 1 such as a TV camera is
It is stored in the image memory 2 and used for image processing.

Under the control of the control unit 4, the image processing unit 3 performs high-speed basic image processing such as edge detection on the grayscale image stored in the image memory 2 to detect the outline of the object in the grayscale image. The component is detected.

With respect to the outline component detected in this way, the control unit 4
Performs the image processing as shown in FIG. 4, for example, and detects the ellipse formed by the outlines in the grayscale image.

The detection result and the like are displayed on the display unit 5.

Now, the flow of the ellipse detection process shown in FIG. 4 will be described. First, various data required for the arithmetic process are initialized (step a). Then, an image is input from the image input unit 1 to the image memory 2 (step b).

Then, for example, the image data of the input grayscale image stored in the image memory 2 is differentiated, and only the pixel data having a certain differential value or more is extracted as a component of the outline of the object (step c). Then, the outline is segmented into a straight line portion and a curved line portion from the curvature of the line segment,
Label each segment (step d).
For each segment, for example, for a straight line part, describe the segment with the coordinate values of its two end points, and for a curved part, the two end points and the fold line approximation of the curve. A segment is described as a series of each coordinate value of a bending point.

After such processing, a combination of two segments that may form an ellipse among the above segments is extracted (step e). That is, the segment forming the straight line portion is excluded, and two combinations of the segment forming the curved line portion are extracted.

Then, with respect to the above two segments, it is first verified whether or not the chords (straight lines connecting both end points of the curved segment) intersect with each other (step f). This verification of the intersection of the strings is performed, for example, as shown in FIG.
A linear equation of straight lines m1 and m2 connecting the respective end points of l2 is obtained, and the coordinates of the intersection point x are calculated. Then, it is carried out by checking whether or not the intersections exist on the straight lines (strings) m1 and m2, respectively.

If it is recognized that the intersection point x exists on the straight lines (chords) m1 and m2, it is determined that the two segments l1 and l2 do not form an ellipse.

If the above test is passed, then each of the strings m1 and m
The straight line n connecting the midpoints of 2 is the segment (arc)
It is verified whether or not it intersects with l1 and l2 (step g).

In this test, for example, a linear equation of a straight line n is obtained as shown in FIG. 6, and the intersections y1, y2, ... Yn of the linear equations and the polygonal lines of the segments 11 and 12 described by the polygonal line approximation as described above. Each coordinate value of is calculated. Then, it is carried out by checking whether or not each of these intersections y1, y2, ... Yn exists on the straight line n.

If even one of the above intersections exists on the straight line n, it is determined that the property of the ellipse is not satisfied.

By such two condition determinations, it is verified whether or not the two segments selected as described above may form an ellipse.

After that, the two segments l1 and l2 that have undergone the above-mentioned test
Then, the elliptic model is estimated (step h). For the estimation of this ellipse model, for example, assuming that the ellipse has a form of a quadratic equation having two variables of x and y as parameters, first, the quadratic equation is applied to the coordinate series describing each of the segments 11 and 12 by the least square method. Find the coefficients of the equation. Then, it is estimated by obtaining the parameters of the ellipse model (the center coordinates, the lengths of the long axis and the short axis, and the inclination) from this coefficient.

Then, the elliptic model is generated according to the above parameters, and an elliptic mask centered on the elliptic model is generated (step i). The generation of the ellipse model is performed by expressing the ellipse model by a polar coordinate equation and setting the angle of 0-3.
The coordinate value when the temperature is changed to 60 ° C. is set in the image memory 2
It is done by writing on the top. Further, the elliptic mask is obtained as a ring-shaped image mask in which the elliptic model is logically filtered and the elliptic model is thickened by several bits.

After obtaining the elliptic mask in this way, it is checked whether or not each of the above-mentioned segments is included in this elliptic mask, and the reliability of the elliptic model is verified (step j). In this verification, for example, as shown in FIG. 7, each segment labeled as shown in FIG. 7A corresponds to the label number (4), (5) 2 shown in FIG.
This is performed by checking whether or not it exists in the annular ellipse mask shown in FIG. 7C estimated from one segment and obtaining the histogram as shown in FIG.

Then, the number of the label whose frequency is not zero is regarded as the segment included in the elliptic mask, and the number is obtained. The number of segments and the label number thus obtained are stored in correspondence with the estimated elliptic model.

It is determined whether or not the above processing has been performed for all pairs of two segments that may form an ellipse in the segment (step k), and the above steps e to step until all the processing is completed. The processes up to j are repeatedly executed.

Next, among the elliptic models estimated as described above, the label competing property of the segments constituting the model is examined. If there are competing labels, the one having a larger number of segments constituting the model, that is, the one having a larger number of segments included in the elliptical model is extracted as having higher reliability (step l).

By the above processing, only the highly reliable elliptic model estimated from the segment is selected, and the segment included in the elliptic model is detected as the elliptical component forming the outline of the object in the grayscale image. Will be done.

Thus, according to this method, unlike the conventional method, the elliptical component can be detected easily, efficiently, and reliably from the line segment forming the outline of the object in the grayscale image. . Further, the elliptic component can be effectively detected from the broken line segment. Therefore, when applied to the image recognition processing, a great effect can be obtained.

The present invention is not limited to the above embodiment. For example, in the procedure of detecting the segment of the ellipse candidate,
If the information such as the size of the ellipse to be detected and its position is used in advance, the processing efficiency can be further improved. Further, by using information such as the positional relationship between the segments and the ratio of the length of the segment to the perimeter of the ellipse, the reliability of the elliptic model with higher reliability can be verified.

It is also possible to verify the reliability of the ellipse model by utilizing the positional relationship with another characteristic figure, for example, by utilizing the condition that parallel lines are in contact with the ellipse. Furthermore,
Of course, it is also possible to modify the estimated ellipse model using other segments included in the ellipse mask. In addition, the present invention can be variously modified and implemented without departing from the scope of the invention.

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a diagram for explaining the property of an ellipse which is the basis of the present invention, FIG. 2 is a diagram showing the basic concept of the present invention, and FIG. 3 is an embodiment of the present invention. FIG. 4 is a schematic configuration diagram of the device, FIG. 4 is a diagram showing a flow of an ellipse detection process in the embodiment device, and FIG.
Fig. 6 is a diagram showing an example of verification of intersection of two strings, Fig. 6 is a diagram showing an example of verification of intersection of a straight line connecting two strings in an ellipse and a segment (arc), and Fig. 7 is extraction of segments included in an ellipse model. It is a figure which shows an example of a process. 1 ... Image input section, 2 ... Image memory, 3 ... Image processing section, 4 ... Control section, 5 ... Display section.

Claims (1)

[Claims] 1. A means for extracting a contour line of an object from a grayscale image, a means for dividing the extracted contour line into segments of a straight line portion and a curved portion, and labeling each segment. For two, test whether the strings of each segment intersect each other,
If they do not intersect, test whether the straight line connecting the arbitrary points of the above chord intersects each segment, and if they do not intersect, use the set of possible ellipse segments as the set of possible ellipse segments. Included in this ellipse mask are means for finding a set, means for estimating an ellipse model from this set of segments, means for generating an ellipse mask as an annular mask image obtained by thickening the estimated ellipse model. An ellipse detection apparatus comprising: a unit for inspecting the number of labeled segments to verify the accuracy of the ellipse model.