基于投影降维的激光雷达快速目标识别  

Fast three dimensional LIDAR target recognition based on projection dimension reduction

作　　者：杨文秀[1] 付文兴[2] 周志伟[3] 余巍[1] 马杰[1] 

机构地区：[1]华中科技大学自动化学院多谱信息处理技术国家级重点实验室,湖北武汉430074 [2]北京机电工程研究所,北京100074 [3]天津津航技术物理研究所,天津300192

出　　处：《红外与激光工程》2014年第B12期1-7,共7页Infrared and Laser Engineering

基　　金：一院高校联合创新基金(CALT201303);航天科技创新基金(CASC04);中央高校基本科研业务费(2014GF)

摘　　要：激光雷达可以获得目标的三维形状信息,是复杂地面背景下车辆识别的有效手段。提出了一种基于投影降维的快速激光雷达目标识别方法,对检测后提取的疑似目标三维点云数据进行投影降维,得到数字表面模型(DSM)数据,根据轮廓相似性度量值以及尺寸相似度量值构成的组合识别准则进行目标识别。采用8组地面装甲目标的仿真点云数据进行实验,实验表明算法的目标型号识别率高于90%,实时性远优于传统算法。针对实际应用需求,进一步研究了点云空间分辨率,激光雷达成像系统误差对目标识别的影响。Laser radar can obtain the three-dimensional shape information of objects, which is an effective method to identify vehicle on complex ground. A fast laser radar target recognition method based on projection dimension reduction was proposed in this paper. Firstly, the suspected target area was extracted and tested from the scene. Then the digital surface model（DSM） data was obtained. Finally, the combination of contour similarity measure value and the similarity of size measurement acted as the matching recognition criteria for target recognition. The experiments were performed by using the simulated point clouds data of 8 ground armored targets. The results show that the type recognition rate of this method is above 90%, and the real-time performance is far superior to the traditional algorithms.In view of the practical application, the influences of point cloud spatial resolution and laser radar system noise on target recognition were further studied.

关 键 词：激光雷达 三维目标识别 投影降维 点云 方向估计 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]