Negative Obstacle Detection in Unstructured Environment Based on Multiple LiDARs and Compositional Features

For negative obstacle detection of autonomous land vehicle (ALV) in unstructured environment, a method based on multiple LiDARs and compositional features is proposed. Firstly, a multi-LiDAR installation manner with complementary ability is proposed. Secondly, two methods are presented:a negative obstacle feature point pair detection method with 64-beam LiDAR based on local convexity in amplitude direction and local dense features at up-side of a ditch, and a negative obstacle feature point pair detection method with 32-beam LiDAR based on range jump in radial direction and local dense features at up-side of a ditch. From the view of spatial and temporal fusion of the negative obstacle, a Bayesian rule is adopted to fuse the feature point pairs from multiple sensors and multiple frames. Then the DBSCAN (density-based spatial clustering of applications with noise) algorithm is applied to clustering and filtering the feature point pairs after fusion. Finally, the data are discretized to extract negative obstacle grid. The experimental results show that the proposed method obtains a good performance for detecting negative obstacles in unstructured environment.