Stereo based navigation in unstructured environments

Abstract

In this paper a stereo vision method is described developed for autonomous vehicle navigation in unstructured terrain. The disparity between the stereo image sequences is calculated with subpixel accuracy. Robust distance values are obtained by adding a continuity criterion term in the evaluation criterion and by using confidence measures for the found disparities. From the disparities a height map is calculated. Path planning is based on these height maps avoiding negative and positive obstacles. The viability of the approach is shown from real stereo image sequences of unstructured terrain.