Robust and Real-Time Road Edge Detection with Flexible LiDAR Configuration

Road edge detection is an important component in autonomous driving for along-side scenarios. The road edge is defined as a boundary line where there is a height change from the road surface. However, the traditional LiDAR-based methods are not sensitive enough for small height difference. In addition, the point cloud needs to be provided in a channel-based organization, which means that at most one LiDAR's data can be used for edge detection. In this paper, a novel road edge detection method is proposed to utilize flexible LiDAR configuration, e.g., multiple LiDARs. Specifically, the raw point cloud is first sorted in polar coordinates, and coarse edge points are generated according to height gradient for accurate ground segmentation. The non-ground points are further used to determine the road types based on occupancy map and beam models. The final edge points in the region of interest are extracted by dynamic slide window. Our method is evaluated on unmanned ground vehicle equipped with two Robosense LiDARs, and extensive experiments are conducted to demonstrate the effectiveness of our method.