A cascaded framework for robust traversable region estimation using stereo vision

Traversable region estimation is the fundamental enabler in autonomous navigation. In this paper, we propose a traversable region segmentation algorithm using stereo vision. We address this problem mainly in road scenes for the goal of autonomous driving. Using only geometry information, our approach has the advantages of effectiveness and robustness. The proposed approach is based on a cascaded framework. Given the disparity map of a rectified stereo image pair, projection from disparity space to U-V-disparity space is performed. We compute the U-V-disparity map in a probabilistic manner so that it is insensitive to errors generated by stereo matching. Then, our approach uses the U-V-disparity map to extract the potential regions. Finally, A RANSAC plane fitting on the potential regions is utilized to remove outliers. The results presented show the validity and robustness of our approach using KITTI road dataset.