Stereovision Based Generic Obstacle Detection and Motion Estimation Using V-stxiel Algorithm

Driving environment perception is a critical and fundamental task of Advanced Driver Assistance System (ADAS) and self-driving technology. In object detection, one of the major challenges is how to achieve an effective model from real scene. Although A medium-level representation named stixel-world has achieve accurate detection results for the task of generic obstacle detection, robustness is still a problem and hard for application. To address this issue, this paper proposes a variant of approach named V-stixel that need no free space computation. Compared with the detection method of monocular image, a variety of obstacle types can be detected without the specific object modeling. It is a generic obstacle detection method and needs no previous training step. Finally, we compare this new method with baseline methods on the KITTI stereo dataset. The experimental results show that the accuracy of the detection results is good, the measurement error is robust in the long video sequences and gain a significant speed-wise.