南京大学学报（自然科学版）2025，Vol.61Issue(5)：738-751,14.DOI:10.13232/j.cnki.jnju.2025.05.003
基于改进DBSCAN算法的道路障碍物点云聚类
Point cloud clustering of road obstacles based on improved DBSCAN algorithm
摘要
Abstract
Obstacle detection technology based on road point cloud data is crucial in intelligent transportation systems and autonomous driving.The traditional density-based spatial clustering(DBSCAN)algorithm has poor clustering effect when processing high-dimensional or different density area data due to inefficient distance measurement and difficulty in determining parameter combinations.Therefore,a road obstacle point cloud clustering method based on improved DBSCAN is proposed.Firstly,the isolated kernel function is used to improve the traditional distance measurement method when determining the Eps area,which improves the adaptability and accuracy of DBSCAN clustering for different density areas.Secondly,in view of the shortcomings of the Cheetah Optimizer(CO)in information sharing and iterative updating,a CO optimization algorithm based on timely updating mechanisms and compatible metric strategies(TCCO)is proposed.The real-time update operation ensures that the excellent information of each iteration is communicated and shared in time,and the elimination mechanism is optimized based on non-dominated sorting and crowding distance during global update to balance the global search and local development capabilities,thereby improving the convergence speed and accuracy.Finally,the Eps field is improved by using the isolation metric,and the DBSCAN clustering is optimized by using TCCO to adaptively determine the parameters,thereby improving the clustering accuracy and efficiency.The simulation results show that the proposed TCCO-DBSCAN algorithm has significantly improved F-Measure,ARI and NMI indicators compared with CO-DBSCAN,SSA-DBSCAN,DBSCAN,and KMC methods,and has better clustering accuracy.The experimental verification of obstacle clustering of lidar point cloud data shows that TCCO-DBSCAN can effectively adapt to the changes in point cloud data density,obtain better road obstacle clustering effects,and provide support for obstacle detection in assisted driving.关键词
DBSCAN聚类/孤立核函数/改进猎豹优化算法/障碍物点云聚类Key words
DBSCAN clustering/isolated kernel function/improved cheetah optimization algorithm/obstacle point cloud clustering分类
信息技术与安全科学引用本文复制引用
吴超凡,黄鹤,贾睿,杨澜,王会峰,高涛..基于改进DBSCAN算法的道路障碍物点云聚类[J].南京大学学报（自然科学版）,2025,61(5):738-751,14.基金项目
国家自然科学基金(52472446),中央高校基本科研业务费(300102325501),陕西省留学人员科技活动择优资助项目(2023001) （52472446）
