基于激光雷达的无人驾驶障碍物检测和跟踪

针对激光雷达动态障碍物检测与跟踪过程中聚类适应性差、实时性低和跟踪准确度不高等问题，提出一种自适应的密度聚类算法和多特征数据关联方法，分别用于检测和跟踪. 首先，对激光雷达采集的点云进行路沿检测、感兴趣区域提取和地面分割等预处理，去除无关点云；然后，基于自适应的密度聚类算法对非地面的点云进行聚类，完成障碍物点云检测；最后，利用加权多特征数据关联算法结合卡尔曼滤波器实现对动态障碍物跟踪. 通过实验表明:本算法能够根据10 Hz的激光雷达数据实现对障碍物准确、稳定的检测和跟踪，且聚类时间缩短32%. 

Obstacle Detection and Tracking for Driverless Cars Based on Lidar

To improve clustering adaptability, real-time performance, and tracking accuracy in the process of dynamic obstacle detection and tracking with lidar, an adaptive density clustering algorithm and a multi-feature data association method are developed for detection and tracking respectively. Firstly, the point cloud collected by the lidar is pre-processed such as curb detection, area-of-interest extraction and ground segmentation to remove irrelevant point clouds. Then the non-ground point clouds are clustered according to the adaptive clustering algorithm to complete the obstacle point cloud detection. Finally, a weighted multi-feature data association algorithm combined with Kalman filter is used to track dynamic obstacles. The experiments show that the entire detection process with the proposed methods can accurately and stably detect and track the 10 Hz lidar data, and shorten the clustering time by 32%.