Driving space detection by combining V-disparity and C-velocity

Abstract

This paper deals with road plane detection by image analysis in the context of automatic driver assistance systems. In this context, free navigable space detection is a very important step for any navigation and obstacle detection system. We propose a low-level combination of two main visual processes: stereovision and motion. We define a common representation that allows simple projections of stereo information to easy-interpretable features in a ”motion” space. We chose to combine two robust cumulative techniques: the stereo-based approach V-disparity and the motion-based approach C-velocity. The combination requires the definition of a common formalism. Results on synthetic image sequences and on KITTI database images reveal that our approach is more efficient than a higher level combination method. We show that it is possible, using no prior knowledge nor any calibration, to improve detection by a low cost method that exploits only image processing and a very simple stereo and motion combination.