Multi-cue road boundary detection using stereo vision

Abstract

In this paper, we propose a multi-cue fusion approach to detect the road boundary using stereo vision, which fits road boundary with a few edge points. Firstly boundary areas are determined in accordance with the normal vector information. Based on the cues of normal vector, height and color in the boundary area, three Bayes models are established respectively. Then the Naive Bayes framework could provide the confidence level of each point in the boundary area, which fuses three kinds of cues. In each boundary area, the point with the highest confidence level would be output. Finally, the support vector regression (SVR) method fits curb curves according to the correct edge points. Extensive experiments tested on KITTIRoad benchmark demonstrate that our method reaches the state-of-the-art.