JPH0760459B2 - Corner detector - Google Patents

Description

TECHNICAL FIELD The present invention relates to a corner detecting device for detecting a corner portion of a rectangular object.

(B) Conventional Technology In recent years, an inspection device for detecting the arrangement state of rectangular chip components arranged on a circuit board is being developed. In such an inspection device, an actual component position is detected by irradiating the circuit substrate with light and capturing an image of the substrate surface with a camera or the like, and detecting a corner portion of a rectangular chip component. . By the way, as a typical method for detecting a rectangular object corner such as a chip part, there is a binary image template matching method. This method binarizes the original image (input image) as shown in FIG. 19 and compares it with the binary template image that it has in advance.
This is to select the points that can be optimally matched. There is also a grayscale template matching method that has a grayscale template image and performs matching with the input original image.

C) Problems to be solved by the invention The binary image template matching method described above is effective in an image that is easily binarized because the difference in density between the corner portion and the background portion of the object to be detected is large, but When the density difference between the corner and the background is small, corner detection is often impossible or erroneous. Further, although the grayscale template matching method eliminates the problem of determining the binarization level, the method of obtaining the grayscale template image needs to be created by referring to the actual camera input image, which complicates the work. Moreover, the time required for matching is far longer than that of the binary image template matching method.
There is a problem that the inspection speed becomes slow.

Further, in any of the above methods, if the density unevenness of the input image due to the fluctuation of the illuminance is large, the corner detection becomes more inaccurate.

D) Means for solving the problems The present invention has been made in view of the above points,
Edge detection means for detecting edge information about each direction by detecting brightness information differences in a plurality of directions from the input image, and edge detection information for obtaining the edge image information, and each of the edge image information for each direction are shifted by the data shift means. After that, the inter-image product of these edge image information is obtained, and a corner candidate point extracting means for extracting a product having a predetermined value or more as a candidate point of the corner portion on the input image, and the candidate point information of the corner portion are registered in advance. And an extraction unit that extracts a corner point from the candidate points of the corner portion by comparing the information indicating the shape of the input image.

(E) Action Candidate points for corners are detected by overlapping edge information in each direction, and corners are extracted from these candidate points, so corners can be detected accurately and quickly. .

F) Example FIG. 1 shows an example of the corner detecting device of the present invention.
(1) is an image capturing unit that captures an image showing the layout of chip components from the printed circuit board, specifically CC
D cameras are used. (2) is an image memory for storing the image thus captured, (3) is a method for calculating the change in luminance around each pixel in the image memory (2) in the vertical and horizontal directions, and for each direction as edge information. An edge extracting means for obtaining the edge degree of each of the vertical and horizontal directions as shown in Japanese Patent Application No. 61-222586, and the edge degree in each of the vertical and horizontal directions is detected. It It is desirable that the operator is a vertical or horizontal operator having anisotropy of 5 × 3, 3 × 5 or more as shown in FIGS. 2 and 3. (4) is the edge of the corresponding pixel by overlapping the center (indicated by diagonal lines in the figure) of the maximum value operator in the vertical and horizontal directions of 5 × 1 and 1 × 5 as shown in FIGS. 4 and 5 with each pixel. A maximum value filter that makes the degree the maximum of the edge degrees of the surrounding pixels covered by this operation,
(5) is the maximum when the centers (indicated by diagonal lines in the figure) of the 1 × 3 and 3 × 1 vertical and horizontal thinning operators as shown in FIGS. 6 and 7 are overlapped on each pixel. With respect to the edge information after the value filter processing, a thinning filter in which only the edge degree of the pixel is larger than the edge degree of the surrounding pixels covered by the thinning operator and the others are set to 0 Thus, the edge portion is narrowed down. (6) is a shift means for shifting the thinned vertical and horizontal edge information. By shifting the vertical and horizontal edge information by a predetermined amount by the shift means, a more effective corner candidate point is obtained. The edge information to be obtained can be obtained. For example, when detecting the upper left corner, the horizontal edge information is shifted to the left and the vertical edge information is shifted to the upper side. (7) is an arithmetic means for superposing (multiplying) the edge degrees in the vertical direction and the horizontal direction thus shifted for each pixel, and the operation result is calculated as the corner degree of each pixel. (8) is a predetermined level of the corner degree calculated by the calculating means (7) (for example, 50% of the maximum corner degree).
The above-mentioned) is a corner candidate point extracting means for selecting the above, and the candidate points are extracted for each of the four corners of the chip component. (9) is an optimum corner point extracting means for extracting the optimum four corners from the extracted candidate points of the corner portion, which is performed by matching with the shape data of the chip part registered in advance. Specifically, the coordinates of the master corner are (Xi, Yi) and the coordinates of the candidate points of the corner are (xi,
yi) (i = 1, 2, 3, 4), the coordinates of the candidate points of the corner when rotating and translating With respect to the above, the set of (xi, yi) may be obtained so that the residual D = Σ (x′i−Xi) + Σ (y′i−Yi). In order to obtain this, x′i = axi−byi + e y′i = bxi + byi + f is obtained from the above equation, and this is substituted into the equation. From the minimum condition Is obtained, and by substituting this into the expression, From the minimum condition By substituting this into an expression And the candidate point (xi, yi) that minimizes this is taken as the corner point. In the above calculation formula, Σ is performed from i = 1 to 4.

In such a corner detecting device, when an image of a chip part as shown in FIG. 8 is captured by the image capturing section (1) from the printed circuit board while illuminating from above and stored in the image memory (2), The edge extracting means (3) is provided with a window (W) at each corner, and calculates the difference in the horizontal direction and the vertical direction in each window for each pixel. FIG. 10 shows the difference information (edge degree) in the horizontal and vertical directions with respect to the upper left corner obtained in this way. Here, it is assumed that the edge degree is higher in the order of black, a shaded portion, and white. At this time, similar differences are calculated for the other three corners, but only the processing for the upper left corner will be described below.
Next, with respect to such edge degree information, the maximum value operator (4) is applied to the maximum value operator to correct gaps (A), (A), ... Edge degree information in the horizontal and vertical directions of continuous lines is formed as shown in FIGS. The edge degree information processed through the maximum value filter (4) is subsequently thinned by a thinning filter (5). As described above, the thinning operator is used to maintain the edge degree only for the pixels having the edge degree higher than the surrounding edge degree, and the others have the edge degree of zero. By such an operation, the horizontal and vertical edge degree information becomes as shown in FIGS. 13 and 14. The horizontal and vertical edge degree information thus obtained is subjected to shift processing by the shift means (6) as shown in FIGS. 15 and 16. In this case, since the image processing is performed for the upper left corner, the shift direction is to shift the horizontal edge degree information to the left and the vertical direction edge degree information to the upper side. The edge degree information of is shifted to the right and the edge degree information in the vertical direction is shifted to the lower side. After the shift, the edge degree in the horizontal direction and the edge degree in the vertical direction are overlapped (multiplied) for each pixel by the calculating means (7) as edge information, and among them, those having a predetermined value or more are detected by the corner candidate point extracting means (8). It is selected as a candidate point for the corner section as shown in Fig. 17. Such candidate points of the corner portion are performed for each of the four corners of the part by the same operation as described above, and among the candidate points thus extracted, the most appropriate four corners are the optimum corner point detecting means. Is selected by.

In the present embodiment, the point of detecting four corners of the part has been described, but it is also possible to confirm the position of the part by detecting two corners or three corners.

G) Effect of the Invention As described above, the corner detecting device of the present invention detects the edge information regarding each direction by detecting the luminance information differences regarding a plurality of directions from the input image, and the edge detecting means for obtaining edge image information, After each edge image information regarding the direction is shifted by the data shift means, an inter-image product of these edge image information is obtained, and a corner candidate point which is a predetermined value or more is extracted as a candidate point of the corner portion on the input image. It has an extracting means and an extracting means for comparing the candidate point information of the corner portion with the information indicating the shape of the input image registered in advance and extracting the corner point from the candidate point of the corner portion. Therefore, even if noise is mixed in the input image information, the most suitable corner point combination is extracted after the candidate points of the corner part are accurately detected, and the position of the part is detected. Can be made more accurate.

[Brief description of drawings]

FIG. 1 is a block diagram of the corner detecting device of the present invention, FIGS. 2 to 7 are schematic state diagrams showing an operator of image processing, FIG. 8 is a state diagram when a window is provided on a captured image, and FIG. Figures 17 to 17 are schematic diagrams of the state of the image processing process by the apparatus of the present invention, FIG. 18 is a schematic diagram showing the distribution of corner candidate points, and FIG. 19 is the state when the binary template image is superimposed on the original image. It is a schematic diagram. (1) …… Image capture unit, (2) …… Image memory,
(3) ... edge extraction means, (4) ... maximum value filter, (5) ... thinning filter, (6) ... computing means,
(8) ... Corner candidate point extracting means, (9) ... Optimal corner point extracting means.

Claims (1)

[Claims] 1. An edge detection unit for detecting edge information about each direction by detecting brightness information differences about a plurality of directions from an input image, and a data shift for each of the edge image information about each direction. After the shift processing by the means, the inter-image product of these edge image information is obtained, and a corner candidate point extracting means for extracting a product having a predetermined value or more as a candidate point of the corner portion on the input image, and a candidate point of the corner portion. A corner detecting device, comprising: an extracting unit that compares information and information indicating a shape of an input image registered in advance to extract a corner point from a candidate point of the corner portion.