Abstract:

For reducing the false detection and miss detection in the process of curb detection， a novel curb detection and tracking method is proposed with 3D-LIDAR as sensor. Firstly， the point cloud is preprocessed， and a distance-based filter is used to filter the interference points in the original point cloud， that affect feature extraction， are filtered by a distance-based filter to enhance the extraction accuracy of curb points. For the filtered point cloud， the ground segmentation method with ground plane segment-wise fitting is used to extract the ground point cloud. Then， an adaptive multi-feature fusion algorithm for curb point extraction is designed by using the spatial features of curbs i.e. height difference， smoothness and angle threshold. Next， aiming at the problem of partial curb loss caused by obstacles， the Rao-Blackwellized particle filter tracker is used to track and predict the curb points. Finally， the method is applied to the multi-condition experiments of the unmanned sanitation vehicle， and the results show that the method can accurately detect the road boundary information， and effectively reduce the false detection and missing detection of curb points.

Key words: curb detection, curb tracking, ground segmentation, 3D-LIDAR