基于改进DBSCAN算法的激光雷达目标物检测方法

A Study of Laser Radar Object Detection Based on Improved DBSCAN Algorithm

传统的DBSCAN聚类算法是基于密度的聚类算法,原始算法在搜索精度和搜索效率上存在一定的局限性。基于LUX4线激光雷达数据点的点云特点,结合DBSCAN算法存在的不足与路面目标物的实际情况,提出了1种基于改进的DBSCAN聚类算法,选取4个代表点取代对所有点的搜索和改进搜索半径使其随扫描的距离而变化的方法,实现激光雷达目标物的快速、准确检测。通过改进DBSCAN算法对雷达数据进行去噪声和聚类处理,根据检测物在激光雷达探测中的形状特征模型进行形状匹配。实验结果表明该改进算法能较好的识别出目标物,行人检测率由原始算法的61.90%提高到了80.95%,搜索时间较原始算法缩短了44.7%,解决了原始算法精度低、搜索慢的缺点。

The original DBSCAN has some limitations in the search accuracy and efficiency.The paper proposed a method of object detection using laser radar based on improved and fast DBSCAN clustering algorithm by taking into consideration of the characteristics of the data collected by LUX 4-line laser radar,the deficiency of original DBSCAN algorithm and the actual situation of road surface objects.The proposed method only searches 4represented points instead of all the points and improves the searching radius that changes by following the scanning distance.The noises are removed and the laser radar points are clustered by improved DBSCAN algorithm.The targets are matched with the shape models of laser radar.The results show that the improved algorithm can accurately identify the objects.Pedestrian detection rate increases from 61.90% of the original algorithm to 80.95%,and the searching time comparing with the original algorithm is shortened 44.70%.The study results clearly indicate that the improved algorithm has an advantage of high precision and fast searching.