A framework for detecting road curb on-line under various road conditions

Quick and accurate understanding of the ambient environment is critical in the development of the intelligent vehicle. As the most important element consisting the ambient environment, road curb detecting is a fundamental and vital work for the development of intelligent vehicle. In this paper, a framework was presented to detect the road quickly and robustly using the velodyne lidar, which provides a massive and unstructured information of the environment in the format of point cloud. Three-sigma rule in the field of error processing and reliability theory was adopted to detect road curb which adapts various road condition. After then, the point cloud was grided and outlier points were removed. Lastly, by utilizing the road continuous information, a search method consists of eight search templates was utilized to search the road curb automatically. The framework we presented has been proved to be able to detect straight or curve road curbs robustly and quickly in most road conditions automatically. Compared to the former works, this paper detects the the road curb depending on a adaptive threshold, which means it can be adaptive to different road conditions. Besides, the search model is adaptive to the changing of the road curb direction, which improve the search speed and increase the road curb detection average accuracy.