JPH0636196B2 - Obstacle detection device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION (Industrial field of application) The present invention relates to obstacles mounted on a moving body such as an autonomous mobile robot that performs maintenance / inspection of a nuclear power generation facility, assembly work in outer space, and the like. The present invention relates to an object detection device.

(Prior Art) An autonomous mobile robot that performs various tasks on behalf of human beings is required in places where it is difficult for humans to reach, such as nuclear power generation facilities and outer space. In order to safely move these robots to the work place, the function of detecting obstacles in the moving environment is essential.

In order to detect such obstacles, proximity sensors using ultrasonic waves have been used in the past, but in recent years, mobile robots have become smaller and lighter in visual sensors and image processing devices have become faster. Attempts to mount an ITV camera on the computer and perform more advanced information processing have become popular.

At this time, the entire image obtained by the ITV camera is processed.
Recognizing and repeating the obstacle detection process is inefficient and impractical. Therefore, a stereo image is acquired by using two ITV cameras, edges are detected from the stereo images, and then the left and right images corresponding to the distance (obstacle detection distance) to the space where the robot is going to move from now on are detected. An edge having a shift (parallax) is extracted as a matching edge,
The method of cutting out the obstacle area by combining these matching edges is M. Watanabe et, al., “Obstacle Detec
tion Method with Stereo Vision ", Proc.of 5th SCIA, v
ol.1, pp325-334 (1987).

However, although this method can easily extract the edges, in order to combine the matching edges, it is necessary to obtain the connection relationship between the edges, and the processing takes a lot of time.

Further, when the obstacle has a curved edge shape like a human, for example, when the edge is discontinuously extracted, the collation edges cannot be combined, and the obstacle cannot be detected. .

(Problems to be solved by the invention) Conventionally, an edge is detected from a stereo image, and an edge having a parallax corresponding to an obstacle detection distance is extracted as a matching edge and combined to cut out an obstacle area. In the technology, there is a problem that it takes a long processing time to combine the matching edges, and it is difficult to detect an obstacle due to a break in the extracted edges.

It is an object of the present invention to provide an obstacle detecting device capable of surely detecting an obstacle without combining matching edges obtained from edges extracted from a stereo image.

[Configuration of the Invention] (Means for Solving the Problems) An obstacle detection device of the present invention is a parallax corresponding to a predetermined obstacle detection distance among edges extracted from each of a pair of left and right stereo images. Only the edges are overlapped with each other, and the similarity of neighboring areas of the overlapped edges is calculated. Among the calculated similarities, the edge with the highest similarity is extracted as a matching edge, and one of the stereo images is an obstacle candidate. It is divided into regions, the inclusion relationship is determined by comparing the position of the obstacle candidate region in the image with the position of the matching edge in the image, and the obstacle candidate region containing the matching edge is determined as an obstacle. It is designed to be detected.

There are various possible processes for dividing one of the stereo images into obstacle candidate regions. Specifically, for example, a density histogram curve is obtained for one of the stereo images, and a minimum value pair adjacent to the maximum value exceeding the threshold value in the curve is formed. The corresponding density area may be the obstacle candidate area.

(Operation) In the present invention, since the obstacle candidate region including the matching edge is detected as an obstacle, the processing for combining the matching edges is not required, the processing time is shortened, and the reliability of obstacle detection is high. Becomes higher.

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

FIG. 1 is a block diagram showing the overall configuration of an obstacle detection device according to an embodiment of the present invention. In the figure, a stereo image input unit 1 is attached to a moving body (for example, an autonomous robot) and captures a stereo image of the moving environment of the moving body, that is, a pair of left and right images. The obtained stereo image is input to the edge extraction unit 2, and the edges in each image are extracted. The extracted edges are input to the edge matching unit 3, and only the edges having the parallax corresponding to the obstacle detection distance are extracted as the matching edges.

On the other hand, either one of the stereo images from the stereo image input unit 1 is further input to the obstacle candidate region dividing unit 4. The obstacle candidate area dividing unit 4 divides an area in the input image in which an obstacle may exist and outputs the divided area. The information of the obstacle candidate area obtained by the obstacle candidate area dividing unit 4 and the edge matching unit 3
The information on the matching edge extracted by is input to the obstacle detecting unit 5, and the inclusion is checked by checking the inclusion relationship between the obstacle candidate region and the matching edge.

Next, a detailed configuration of each part in FIG. 1 will be described.

FIG. 2 shows a configuration example of the stereo image input unit 1. The stereo images captured by the two ITV cameras 11 and 12 are digitized by the A / D converters 13 and 14, and then the ITV interfaces 15 and 16 and the image bus 1 are used.
9 and stored in the digital image memories 17 and 18. A control bus 20 is also connected to the image memories 17 and 18. It is desirable that the optical axes of the ITV cameras 11 and 12 are set to be parallel in order to simplify the matching process in the edge matching unit 3.

FIG. 3 shows a configuration example of the edge extraction unit 2. The image stored in the digital image memories 17 and 18 of FIG.
After the second differential type filtering is applied, the binarization circuit 32
Is converted into a binary image. This binary image is a thinning circuit 3
In step 3, the binary edge image having the line width “1” is shaped. This binary edge image is input to the labeling circuit 32, and the length of each edge is measured. Then, the noise removing circuit 35 removes the short edge length measured by the labeling circuit 34 as noise, and creates an edge image for stereo matching.

FIG. 4 shows a configuration example of the edge matching unit 3. For edge matching, the edge image extracted by the edge extraction unit 2 is traced with respect to the stereo image previously obtained by the stereo image input unit 1, and the edge position and the connection relationship between the edges are checked.
Edge segment tables 41 and 42 in which attribute information such as end points and bending points of each edge are described are created.

The parallax setting circuit 43 sets the parallax S corresponding to the distance L (obstacle detection distance) to the space where the robot is about to travel. That is, the ITV cameras 11 and 12
Between the obstacle detection distance L and the parallax S when the optical axes are aligned, there is a relation of S = ε · a · / L ε: conversion ratio a: interval between ITV cameras 11 and 12: focal length Since ε and a are constants determined by the configuration of the stereo camera, the parallax S can be obtained if the obstacle detection distance L is determined.

The parallax shift circuit 44 shifts the left and right edge images obtained by the edge extraction unit 2 from each other by the parallax S and then superimposes them. The similarity calculation circuit 45 calculates the similarity of the neighboring areas of the overlapping edge images. Then, the matching edge extraction circuit 46 extracts an edge image having a high degree of similarity as a matching edge, and further, the edge segment table 4
1 and 42, the positions of the matching edges on the image are obtained and stored in the matching edge table 47.

Since the edge extracting unit 2 and the edge collating unit 3 are described in detail in the above-mentioned publicly known documents, further detailed description will be omitted.

FIG. 5 shows a configuration example of the obstacle candidate area dividing unit 4. In this example, by using the heuristic that "when an object exists in the scene, a peak corresponding to the brightness of the object appears on the density histogram curve", the density histogram measurement circuit 51 first uses the entire one of the stereo images. A density histogram curve for is obtained. Next, after the smoothness of the histogram curve is corrected by the smoothing circuit 52, the maximum value of the corrected histogram curve and the density value in which the number of pixels is greater than or equal to the threshold value are obtained by the extreme value detection circuit 53, and the maximum value is obtained. Minimum value pair adjacent to (maximum value before and after 2
The density values at one minimum value) are sequentially registered in the density conversion table 54.

FIG. 6 shows an example of the density histogram curve. In this case, since the density histogram curve has two maximum values P1 and P2, the density values at the adjacent minimum value pairs L1, L2 and L3, L4 are registered in the density conversion table 54. Then, according to the contents of the density conversion table 54 in the density conversion circuit 55, the densities of the areas in the same density range are replaced with a single density (multi-valued), and the obstacle candidates divided by the area description circuit 56 are further divided. The position of the area in the image is measured, and the position information is stored in the area table 57. In the example of FIG. 6, one of the stereo images input to the obstacle candidate area dividing unit 4 is first divided into areas A, B, C, D and E. Of these areas A to E, the minimum value pair L1, L
The areas B and D sandwiched by 2 and L3 and L4 are finally divided as obstacle candidate areas, and the remaining areas A, C and E are recognized as the background portion.

FIG. 7 shows a configuration example of the obstacle detection unit 5. In the range comparison circuit 71, the matching edges stored in the matching edge table 47 in FIG. 4 and the area table 57 in FIG.
By comparing the position of the obstacle candidate area stored in the image with the position in the image, the inclusion relation between the two is checked, and the obstacle candidate area including the matching edge is selected. Then, from the selected obstacle candidate area, the density conversion table 5 in FIG.
4 and the contents of the density conversion circuit 55, the corresponding portion in the image is extracted as an obstacle. Actually, the matching edge is not always on the contour of the area due to the influence of the low-pass filter used in the binarization circuit 32 and the noise removal circuit 35 in FIG. 3, so the area is expanded outward by a fixed width. It is desirable to check the inclusion relationship between the expanded area and the matching edge.

FIG. 8 shows how obstacles are extracted from the scene having the density histogram curve shown in FIG. 6, a to j are matching edges, and regions B'and D'shown by broken lines are obstacle candidate regions, respectively. It is an expansion area in which B and D are expanded outward by a fixed width. In the example shown in the figure, the matching edges a to g are added to the extended area of B '.
Is included, the obstacle candidate region B corresponding to this extension region B ′ is extracted as an obstacle. The other matching edges h to j are included in the area A, but since the area A is not an obstacle candidate area, these matching edges h to j are determined to be false matching in the background portion.

[Effects of the Invention] According to the present invention, by detecting an obstacle candidate region including a matching edge as an obstacle, it is possible to detect an obstacle without performing a process of combining the matching edges, and the processing time is shortened. Therefore, the reliability of obstacle detection can be improved.

[Brief description of drawings]

FIG. 1 is a block diagram showing a schematic configuration of an obstacle detection device according to an embodiment of the present invention, FIG. 2 is a diagram showing a configuration example of a stereo image input section in FIG. 1, and FIG. 3 is FIG. FIG. 4 is a diagram showing a configuration example of an edge extraction unit in FIG. 4, FIG. 4 is a diagram showing a configuration example of an edge matching unit in FIG. 1, and FIG. 5 is a diagram showing a configuration example of an obstacle candidate region dividing unit in FIG. FIG. 6 is a diagram showing an example of a density histogram curve for explaining the processing of the obstacle candidate region dividing section of FIG. 5, and FIG.
The figure which shows the structural example of the obstacle detection part in a figure, FIG. 8 is a figure for demonstrating the process of the obstacle detection part of FIG. 1 ... Stereo image input section, 2 ... Edge extraction section, 3 ...
... edge matching unit, 4 ... obstacle candidate region dividing unit, 5 ...
Obstacle detector.

Claims (2)

[Claims] 1. A means for acquiring a stereo image of a moving environment of a moving body, a means for extracting an edge from each of the acquired pair of left and right stereo images, and a pair of the left and right stereo images for which each edge has been extracted, A means for calculating the similarity between adjacent areas of overlapping edges by shifting them by parallax corresponding to the set obstacle detection distance, extracting the edge with a high similarity as a matching edge, and the stereo image A means for dividing one of the obstacle candidate areas and a position of the obstacle candidate area divided by this means in the image and the position of the matching edge in the image are compared to determine the inclusion relation. An obstacle detection device comprising: means for detecting an obstacle candidate area including a matching edge as an obstacle. 2. A means for dividing one of the stereo images into obstacle candidate regions obtains a density histogram curve for one image of the stereo image and forms a minimum value pair adjacent to a maximum value exceeding a threshold value in the curve. 2. The sandwiched density area is used as an obstacle candidate area.
The obstacle detection device described.