Obstacle detection using trinocular stereo cameras without a disparity map

Abstract

We propose an obstacle detection method based on trinocular stereo vision which consist of vertical and horizontal stereo cameras. Vertical edges and horizontal edges are extracted from horizontal stereo images and vertical stereo images. Then the edges are used to generate U-V-disparity maps. A rectangular obstacle has a characteristic that it is represented as a vertical line in V-disparity map and horizontal line in U-disparity map. Therefore an obstacle can be detected by analyzing line shapes in the U-V-disparity maps. Unlike conventional methods, the proposed method does not need to generate a disparity map which requires expensive computational cost. Therefore, the proposed method can operate in real-time.