Obstacle location estimation using planar projection stereopsis method

Abstract

This paper presents a new method for estimating locations of moving obstacles from stereo images without tracking between image frames and without searching corresponding points between stereo images. First, a free-space map in front of a vehicle is created by the planar projection stereopsis method which can easily decide whether or not each point in a stereo image exists on a ground plane. Next, the moving obstacles detected in an image by using the residual error calculated in the process of focus of expansion estimation are projected to the free-space map. In the free space map, obstacle locations are estimated as the intersection between the projected obstacle areas and the contour of a free-space area. Experimental results for real road scenes with pedestrian, bicycle or automobile as moving obstacles show the effectiveness of the proposed method.