针对64线激光雷达数据量大,导致无人自主车的障碍物检测实时性差的问题,提出一种兼顾有效性和实时性的目标检测和分类算法.该算法首先通过多特征多层高度地图分离路面、障碍物和悬挂物;然后采用基于动态距离阈值的网格聚类算法对障碍物进行聚类,并结合相邻两个障碍物的运动状态信息对聚类结果进行修正,提高聚类的准确率;最后使用SVM对障碍物进行检测和分类.实验结果表明:该算法最优识别率达89.77％,耗时约为95 ms,在保证检测和分类准确率的基础上,满足无人自主车在道路行驶时检测障碍物的实时性要求,具有显著的工程实用价值.%In view of the poor real-time performance of unmanned autonomous vehicle in detecting obstacles due to the huge volume of 64-line lidar data, an object detection and classification algorithm with good effectiveness and real-time performance is proposed. The algorithm separates the road, obstacle and suspended object by multi-feature / multi-layer elevation map. Then the grid clustering algorithm based on dynamic distance threshold is used to cluster the obstacles, with the clustering results corrected according to the motion state information of two adjacent obstacles to enhance clustering accuracy. Finally, SVM is adopted to detect and classify obstacles. The experiment results show that the algorithm attains a best identification rate of 89.77% with a duration of 95 ms, meeting the real-time requirements of unmanned vehicle in detecting obstacle on road, while ensuring the accuracy of detection and classification, having a significant engineering application value.
展开▼
