Negative Obstacle Detection in Off-Road Environments Using Sparse 3D LiDAR Point Cloud

Negative obstacle detection is crucial to ensure safe navigation of unmanned ground vehicles (UGV) in off-road environments. Unlike previous approaches developed using high-resolution LiDAR or multi-LiDAR fusion, this paper proposes a negative obstacle detection method applicable to a single low-resolution LiDAR. First, a ground segmentation method is used to segment the point cloud sampled by LiDAR into ground points and obstacle points. Then, the point cloud is serialized by projecting it into a front-view image. Next, combined with the ground segmentation results, the range view image is traversed, and the feature values of each point in the image are extracted and analyzed to obtain negative obstacle detection results. Finally, the detection results of multiple frames are fused using a Bayesian rule to improve the detection robustness. To verify the effectiveness of the proposed method, we conduct a series of experiments in several typical off-road scenarios. The experimental results show that the proposed method can accurately detect negative obstacles with high real-time performance.