JPS6029995B2 - 3D object detection method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a method for detecting a three-dimensional object, and particularly to a method for detecting a three-dimensional object in a visual device of a mobile machine such as a robot that moves in an environment such as a factory or a road.

For example, in the case of a guide dog robot that assists a blind person in walking alone on the street, the device itself determines its position and direction using landmarks placed at street intersections, etc., and maps of the environment such as landmarks built into the machine. By using this information, you can guide the blind person according to his/her request.

At that time, there are various obstacles on the road, so it is necessary to detect them and take the optimal action according to the situation at the time. Obstacles can be broadly classified into moving obstacles and stationary obstacles, and it is also necessary to detect holes, steps, etc. in the ground.
When it comes to moving obstacles, if we use ultrasonic measurement to determine their position, direction of movement, and speed, if someone is walking in front of them in the same direction at a similar speed, the object will move at the same speed. It is relatively easy to take measures such as waiting for the other vehicle to go too far or issuing a warning to the other vehicle if there is a risk of collision even if the vehicle has stopped. . However, when dealing with stationary obstacles, information such as the object's exact location, size, shape, etc. is important, and ultrasonic measurement alone is insufficient, and it becomes necessary to use visual sensors. . As a visual sensor,
Many studies have been published to date that measure the position of three-dimensional objects using multiple cameras and a stereoscopic method of kneading. Generally, the three-dimensional coordinates in space can be easily calculated using the principle of triangulation once the corresponding points on the two-dimensional coordinates of the left and right cameras are determined. The problem is how to determine corresponding points.
Correlation methods, DP methods, and methods using image structure have been proposed for this purpose, but they either cannot uniquely determine corresponding points, or even if they are determined, they require a huge amount of calculation. Therefore, it is not suitable for controlling mobile robots.
For these reasons, it is desired to develop a device or method that can more easily detect three-dimensional objects. This invention was made in view of the above circumstances, and enables accurate detection of three-dimensional objects with simple operations, simplifies information processing, and enables real-time processing. The object of the present invention is to provide an inexpensive three-dimensional object detection method. Corresponding to this purpose, the three-dimensional object detection method of the present invention includes two cameras rotatably attached to a moving body with respect to an axis of three-dimensional coordinates, and one screen of one camera of a specific object. The intersection of the above image and the virtual image on the other screen of the vector of the line of sight corresponding to the specific object of the one camera among the images on the other screen of the other camera of the object. Comparison is measured between a fixed point on the travel path of the moving body and a point at which the moving body is moved in the line of sight direction of the one camera,
Comparing the results of each measurement,
It is characterized by determining corresponding points of viewpoints on dimensional coordinates.

The details of this invention will be explained below with reference to one embodiment.

In FIG. 1, 1 is a detection device.

The detection device 1 includes a mobile vehicle 2. The mobile vehicle 2 is equipped with wheels 3 and can move in any direction, and has a built-in driving device and a control device, as well as a console 7, a cassette tape device 8, a microcomputer 9, an image capture/storage device 11, Cassette tape console control device 1
2, an 0/A converter 13, a servo amplifier 14, and a camera control device 15. The position of the mobile vehicle 2 is measured by a control device. Two television cameras 4a and 4b are attached to the moving vehicle 2. The cameras 4a and 4b are each rotatable about three-dimensional coordinate axes (0-x, Y, Z). The rotation angles of cameras 4a and 4b are measured by a control device. The cameras 4a, 4b are also connected to the control device via separate image devices 5a, 5b (FIGS. 1 and 2), respectively. The image devices 5a and 5b are connected to an image processing device. Images on the image devices 5a and 5b are measured by a control device. The control device includes a calculation device. In the three-dimensional object detection device configured in this way, in the three-dimensional coordinate system ○-XYZ, the camera 4a
The center of is (X,,Y,,Z,), and the center of camera 4b is (
X2, Y2. Z2), and the direction cosine of the center line of each lens is (in,,in,,shi,,),(^2
, Buddha 2, Shi 2).

Then, the presence or absence of the object 10 to be explored and the measurement of its position are performed in the following steps. That is, as shown in FIG. 2, preprocessing is performed on the image on the image device 5a based on the signal from the camera 4a, and an object 10 such as an obstacle and a pattern of interest are captured at the center height of the field of view. Vector V1 (VI1, V12
, V13) can be found (Fig. 1). At this time, the virtual VI
The vector image taken by the camera 4b becomes a straight line on the bride surface of the image device 5b. A corresponding point of the image captured by the right camera 4a exists on this line segment. Therefore, on this line segment, a pattern similar to the pattern of the right camera 4a is selected as a candidate for the corresponding pattern. In this way, camera 4 of virtual vector VI
Since candidate points are searched on the straight line that is the image in b, the search range is narrowed and processing can be simplified. Then, if necessary, capture it in the center window. At that time, for example, by placing the center of gravity at the center, the center of the figure and the center of the camera are aligned. (Figure 3b). In addition, camera 4
b does not necessarily need to be moved, and a vector (in 2, bud 2, shi 2) connecting the candidate point and the center of the camera may be calculated. Next, two cameras 4a captured the central panicle.
, 4b are determined by the following equation. Accordingly, the object 1 from each camera 4a, 4b is
Find the distances r, b, r2b to 0.

Furthermore, the degree of matching between two patterns is determined by, for example, extracting the outline of the image and using a modified cross-correlation function or the like. Let L, t,, etc. be threshold values. If no, discard that pattern. Few.

>L, the mobile vehicle 2 is moved forward by d in the direction of the line of sight of the right camera 4a, as shown in FIG. 4b. If the size of the image of the pattern before moving forward is B', the size of the image after moving forward is B, and the focal length of the lens is f, then the distances to the object are r, m, and the actual size of the object. A can be easily found using the formula. rm=B kotoku, A 1000 harm...■ Next, point the left camera 4b at point K, where the object should be,
Calculate the degree of coincidence again.

If the correspondence is correct, there is an object at point K, and if the correspondence is correct, then L, > L is satisfied, and the relationship of distances is r, m + d - r, b < t2 ( In the case of Figure 4a). On the other hand, if the above relationship is not satisfied, Fig. 4b
Since there is a false correspondence like , we return to the process of searching for another corresponding point candidate. As described above, according to the present invention, it is possible to accurately detect a three-dimensional object with a simple operation by using both camera movement information and moving vehicle movement information. It is possible to obtain an inexpensive three-dimensional object detection method that allows for simplified processing.

[Brief explanation of the drawing]

FIG. 1 is a perspective explanatory diagram of a three-dimensional object detection device according to an embodiment of the present invention, FIG. 2 is an explanatory diagram showing an example of an image plane of an image device, and FIG. 3 is an explanatory diagram of another image plane of the image device. An explanatory diagram showing an example and FIG. 4 are diagrams showing an operation for determining an object. 1...Detection device, 2...Moving vehicle, 4a, 4b...
Camera, 5a, 5b...imaging device. Figure 2 Figure 3 Figure 1 Figure 4

Claims (1)

[Claims] 1 Equipped with two cameras rotatably attached to a moving body about a three-dimensional coordinate axis, an image of a specific object on one screen of one camera, and an image of the object on the other screen of the other camera. The vector of the line of sight corresponding to the specific object of the one camera in the image intersects with the virtual image on the other screen, and the comparison is made between a fixed point on the travel path of the moving body and the vector of the line of sight corresponding to the specific object of the one camera A point at which the moving object is moved in the line of sight direction of the camera is measured, and the results of each measurement are compared to determine a corresponding point of the viewpoint on the two-dimensional coordinates of the two cameras. A method for detecting 3D objects.