Traversable Region and Obstacle Detection Using Tilted-plane-based Stereo Vision

For autonomous mobility of unmanned or tele-operated vehicles, the visual navigation capabilities such as obstacle detection and terrain estimation are essential. This paper presents a robust model-based approach for detecting traversable regions and obstacles even in off-road environments. In our approach, a model-based polynocular stereo, which we call tilted-plane-based stereo, is utilized. We newly define the algorithm by using a group of slant parallel planes to be searched in the spatial disparity space. It is capable of directly measuring height differences in the view field. By checking abrupt jumps in height in the 3D space, traversable regions of the terrain can be effectively estimated. Then, the obstacles belong to the untraversable regions are robustly detected. We show successful results of the proposed method for different obstacles in the real scenes.