A Two-Stage Adaptive Clustering Approach for 3D Point Clouds

In this paper, we propose a simple and efficient method for the 3D point clouds clustering. Emitted from the 3D Lidar sensor that amounted on the top of the vehicle, the point clouds are sparse and disordered, which bring difficulties in the clustering stage. Clustering the points into optional meaningful objects is the primary work in the perception of the autonomous vehicle, whose performance and efficiency will directly affect the subsequent pipeline including recognition, classification and tracking. Focusing on the sparse and disordered characteristics of point clouds and the requirements of our actual scene, we developed a two-stage adaptive method. In the first stage, we use the Euclidean-based method combined with a sliding window to get small subclusters. In the second stage, we use the adaptive DBSCAN algorithm to get the result clusters, which can efficiently avoid the over segmentation problems.