Trinocular Vision Using Edge Continuity and Local Features

In the mechanical recognition of three‐dimensional shape, position and orientation of the object, it is important to obtain information concerning the distance to the object (depth). One of the powerful approaches for this purpose is the binocular vision (stereo method), in which two cameras are used to determine the distance based on the principle of trigonometric measurement. In this method, the three‐dimensional position of a point can easily be determined if the correspondences of feature points are established between the two images. However, the correspondence is not easy if the two images have a large number of feature points. We investigated the correspondence problem, and solved the difficulty by introducing another camera, realizing the trinocular vision and applying a simple geometrical processing. However, when the number of the feature points on the image is increased in this method, the correspondence is sometimes not unique, leading to more than one corresponding point. To solve this problem, the edge continuity and the local property of the image are combined with the geometrical constraints by trinocular vision, arriving at a satisfactory result.