Research on 3D Point Cloud Data Preprocessing and Clustering Algorithm of Obstacles for Intelligent Vehicle

Abstract

Environment perception is the foundation of the intelligent driving system and is a prerequisite for achieving path planning and vehicle control. Among them, obstacle detection is the key to environment perception. In order to solve the problems of difficult-to-distinguish adjacent obstacles and easy-to-split distant obstacles in the traditional obstacle detection algorithm, this study firstly designed a 3D point cloud data filtering algorithm, completed the point cloud data removal of vehicle body points and noise points, and designed the point cloud down-sampling method. Then a ground segmentation method based on the Ray Ground Filter algorithm was designed to solve the under-segmentation problem in ground segmentation, while ensuring real time. Furthermore, an improved DBSCAN (Density-Based Spatial Clustering of Application with Noise) clustering algorithm was proposed, and the L-shaped fitting method was used to complete the 3D bounding box fitting of the point cloud, thus solving the problems that it is difficult to distinguish adjacent obstacles at close distances caused by the fixed parameter thresholds and it is easy for obstacles at long distances to split into multiple obstacles; thus, the real-time performance of the algorithm was improved. Finally, a real vehicle test was conducted, and the test results show that the proposed obstacle detection algorithm in this paper has improved the accuracy by 6.1% and the real-time performance by 13.2% compared with the traditional algorithm.