Point feature-based outdoor SLAM for rural environments with geometric analysis

This paper proposes a point feature-based outdoor SLAM method using only omnidirectional LIDAR. 3D local occupancy grid mapping and ground plane classification are conducted as a pre-process to refine the point cloud. Then uncertain objects are clustered with Euclidean distance. For applications in rural environments, point features are utilized because clusters are extracted from unclear and overlapped objects. To improve matching performance, the similarity of clusters is calculated with a Hausdorff distance and correspondence filtering with the point histogram is implemented. With the correspondence filtering, we can reduce false matches that cannot be removed from the initial matcher and thus improve the SLAM accuracy. The remaining point features are used as landmarks in SLAM, and the effectiveness of the scheme is verified through simulations with the real-world dataset.