分区基于密度的聚类算法在激光雷达行人检测系统中的应用  

The Application of Partitioning-density-based Spatial Clustering of Applications with Noise Algorithm in LIDAR Based Pedestrian Detection System

作　　者：宋柱 付锐[1] 张名芳[1] 刘新雨[1] 

机构地区：[1]长安大学汽车学院,西安710064

出　　处：《科学技术与工程》2017年第18期282-287,共6页Science Technology and Engineering

基　　金：长江学者和创新团队发展计划项目(IRT1286);陕西省自然科学基金(2016JQ5096);中央高校基本科研业务费专项资金(10822151028;310822172001)资助

摘　　要：行人检测过程中原始DBSCAN算法不能正确地对密度不均匀的激光点云聚类,产生错误的聚类结果导致行人检测系统出现误检和漏检。为解决这一问题,基于激光雷达的行人检测系统在原始密度聚类算法DBSCAN的基础上提出了分区DBSCAN算法。该算法将密度不均匀的点云数据划分为若干个密度相对均匀的分区,从而能实现对行人的快速准确检测。实验结果表明原始DBSCAN算法行人检测率为62.47%,使用分区DBSCAN算法的激光雷达行人检测系统行人检测率达到82.21%,相对于原始DBSCAN算法检测精度提高了19.74%;而且在时间消耗上也比原始DBSCAN算法降低了16.22%。In the process of pedestrian detection, original DBSCA Nalgorithm can′t correctly cluster the uneven laser points cloud, and the wrong clustering result will lead to false detection and leak detection.To solve this problem, a partitioning-DBSCA Nalgorithm was proposed based on the original DBSCAN algorithm for pedestrian detection system.The algorithm will divide the uneven points cloud into several relatively homogeneous density partition, which can realize fast and exact detection.Experimental results show that the pedestrian detection rate of original DBSCAN algorithm was 62.47%, and the pedestrian detection rate of partitioning-DBSCAN was 82.21%, which increased by 19.74%;also, the time consumption of our method was reduced by 16.22% than the original DBSCAN algorithm.

关 键 词：分区基于密度的聚类(DBSCAN) 算法 行人检测 激光雷达 聚类 

分 类 号：TP391.75[自动化与计算机技术—计算机应用技术]