Modeling the Ground Plane Transformation for Real-Time Obstacle Detection

摘要

Ground plane obstacle detection is a general problem for mobile robots. A number of different approaches have been investigated in previous research; however, none is suitable for use with an active stereo vision system. In this paper, we present a real-time approach to obstacle detection for an active stereo vision system based on plane transformation. By modeling the ground plane transformation between a stereo image pair captured by a common elevation stereo head, then identifying the related camera intrinsic parameters, we demonstrate that the ground plane transformation can be computed in real-time using the identified parameters and the head feedback state. Hence it is possible to use it to detect ground plane obstacles in real-time. We also show that this formulation of ground plane obstacle detection unifies previous approaches based on predicted ground plane disparities or on a predicted ground plane transformation for stereo cameras with fixed geometry.