基于切片采样和质心距直方图特征的室外大场景三维点云分类  

Three-Dimensional Point Cloud Classification of Large Outdoor Scenes Based on Vertical Slice Sampling and Centroid Distance Histogram

作　　者：佟国峰[1] 杜宪策 李勇[1] 陈槐嵘 张庆春 Tong Guofeng; Du Xiance; Li Yong; Chen Huairong; Zhang Qingchun(College of Information Science and Engineering, Northeastern University, Shenyang, Liaoning 110819, China)

机构地区：[1]东北大学信息科学与工程学院,辽宁沈阳110819

出　　处：《中国激光》2018年第10期150-158,共9页Chinese Journal of Lasers

基　　金：国家自然科学基金(61175031);国家863计划子课题(2012AA041402);国家科技支撑计划子课题(2015BAF13B00-5)

摘　　要：三维(3D)点云数据在智能驾驶、遥感测量和虚拟现实等领域的应用越来越广泛。针对室外大场景,提出了一种兼顾快速性与准确性的三维点云分类算法,该算法首先对原始点云进行离群点去除,并在现有地面滤波算法的基础上,结合点云法向量差信息滤除地面点;然后再使用具有噪声的基于密度(DBSCAN)的聚类算法对非地面点云进行分割,同时针对点云的过分割问题采用了就近融合的策略;再提取出不同物体点云的全局特征,包括垂直方向切片采样直方图和质心距直方图,以及点云的二维投影图像方向梯度直方图(HOG)特征;最后,通过支持向量机(SVM)分类器分类,得到较为精确的三维点云分类结果。实验结果表明:所提算法可以将复杂的室外大场景分类为较为准确的单个物体,并且具有较高的精确率以及召回率;相较于其他算法,所提算法的效率有了较大提高。Three-dimensional （3D） point cloud data are widely used in intelligent driving, remote sensing, and virtual reality. This study presents a 313 point cloud classification algorithm that classifies large outdoor scenes effectively and accurately. First, the algorithm eliminates outliers from the original point cloud. Then, based on the off-the-shelf ground-filtering algorithm, it leverages difference of norms to filter ground points. Then, it uses the density-based spatial clustering of applications with noise （DBSCAN） clustering algorithm to segment non-ground point cloud. The nearest fusion strategy is used to solve the oversegmentation problem of the point cloud. Then, the proposed algorithm extracts global features that represent different objects from the point cloud, including vertical slice sampling and centroid distance histograms, as well as histogram of oriented gradient （HOG） features representing a two-dimensional projected image of the point cloud. Finally, a support vector machine （SVM） classifier is used to obtain the accurate 3D point cloud classification results. The experimental results reveal that the proposed algorithm can classify complex large outdoor scenes into accurate single objects with high accuracy and high recall rate. The proposed algorithm is more efficient compared with other algorithms.

关 键 词：机器视觉 室外大场景 法向量差 点云分割 特征提取 点云分类 

分 类 号：TN958.98[电子电信—信号与信息处理]