A Novel Method for Obstacle Detection in Front of Vehicles Based on the Local Spatial Features of Point Cloud

Obstacle detection is the primary task of Advanced Driving Assistance System (ADAS). However, it very difficult to achieve accurate obstacle in complex traffic scenes. To this end, paper proposes an method based on local spatial features point clouds. Firstly, cloud a superpixel obtained through stereo matching and SLIC image segmentation algorithm. Then, probability corresponding area estimated from feature information plane normal vector point-cloud height, respectively. Finally, results two methods are input into Bayesian framework form probabilities for final decision. In order describe scene efficiently accurately, further transformed multi-layer stixel representation. We carried out experiments KITTI dataset compared several methods. The experimental indicate that proposed has advantages terms its Pixel-wise True Positive Rate (PTPR) False (PFPR), particularly scenes, such as uneven roads.