Obstacle Detection of Unmanned Surface Vehicle Based on Lidar Point Cloud Data

• Published in: OCEANS 2022, Hampton Roads pp. 1 - 8
• Main Authors: Deng, Lihui, Guo, Tingting, Wang, Hongjian, Chi, Zhikang, Wu, Zhiqiang, Yuan, Rubin
• Format: Conference Proceeding
• Language: English
• Published: IEEE 17.10.2022
• Subjects: Clustering algorithms; DBSCAN algorithm; Filtering algorithms; Laser radar; lidar; Navigation; obstacle detection; Point cloud compression; point cloud data; Real-time systems; Three-dimensional displays; unmanned surface vehicle
• DOI: 10.1109/OCEANS47191.2022.9977340

Currently, most algorithms are difficult to adapt to the real sea level scene point cloud target detection, in this paper, a navigation obstacle detection algorithm for unmanned surface vehicle (USV) is proposed, which can detect the obstacles during the navigation of USV and provide real-time and effective obstacle information for USV. The navigation obstacle detection algorithm includes four algorithms: laser radar point cloud preprocessing algorithm, forward looking grid network building method, improved RANSAC filtering algorithm, and improved DBSCAN clustering algorithm. According to the characteristics of the lidar point cloud data, the lidar point cloud preprocessing algorithm is adopted, which is named the multi frame point cloud data processing algorithm. Then the forward looking grid network is established, and the 3D point cloud data is placed on the forward looking grid network for processing, which greatly speeds up the processing speed of point cloud data and improves the real-time performance of obstacle detection. For the point cloud data of lidar that contains many clutter reflections, the point cloud that reflects due to wake waves generated by ship navigation is filtered (filtered), and the improved RANSAC algorithm is used to filter the point cloud to reduce the interference of clutter point cloud data. In this paper, a DBSCAN algorithm with adaptive parameters is proposed to solve the problem of point cloud target clustering, combined with the scanning characteristics of lidar. In this way, it retains the advantages of traditional DBSCAN in short distance point cloud clustering, and can correctly cluster long distance point clouds with good results.