Reliable omnidirectional depth map generation for indoor mobile robot navigation via a single perspective camera

This paper deals with the problem of finding the largest navigable areas around a mobile robot, which is important for navigation and action planning. We propose a method to obtain reliable dense 3D maps using a novel omnidirectional stereo vision system. The vision system is composed of a perspective camera and two hyperbolic mirrors. Once the system has been calibrated and two image points respectively projected by upper and below mirrors are matched, the 3D coordinate of the space point can be acquired by means of triangulation. To achieve the largest reliable dense matching, our method are divided into three steps. First reliable FX-dominant matching; then feature matching and ambiguous removal; finally the remaining points between features are matched using dynamic time warping(DTW) with modified energy functions adapted well to our system. Experiments show that this proposed vision system is feasible as a practical stereo sensor for accurate 3D map generation