机载扫描激光雷达点云形态分析滤波  

Filtering Airborne Lidar Data with Morphological Analysis Methods

作　　者：姚树森[1] 徐集贤 陈思颖[2] 

机构地区：[1]军事医学科学院科技部,北京100071 [2]北京理工大学光电学院,北京100081

出　　处：《北京理工大学学报》2014年第4期401-406,共6页Transactions of Beijing Institute of Technology

摘　　要：研究多种地理形貌下,从激光雷达点云中识别地面点(滤波)并同时实现场景主要形态特征分析的方法.首先采用伪栅格和混合插值技术保证点对点的快速索引和操作;然后在多尺度的开运算框架下,测量场景背景、对象尺度和方向分布,估计滤波参数,进行基于边缘统计的地面点识别;最后基于残余点和初始地面的局部关系进行地面点的二次识别.实验采用国际遥感和摄影测量学会(ISPRS)的标准数据集,结果显示在多种地理形貌下的形态测量准确合理,平均滤波精度达到90%以上;尤其在植被和建筑混合的山坡场景中,效果改善明显.A multi-scale solution was proposed for synchronous implementation of ground point identification （point clouds filtering） and morphological feature analysis, to adapt to various landscapes. First, a pseudo-grid and hybrid-interpolation methods were employed for effective operation of the point clouds. Then, the morphological analysis, including the background estimation, object scale analysis and object direction analysis, was implemented together with an edge-based morphological filter in a same multi-scale framework. The morphological analysis also helps estimate parameters for the filter. Lastly, a second filtering process was followed to identify ground points from residual points, based on the local landscape features. The tests with the benchmark dataset from international society for photogrammetry and remote sensing （ISPRS） show that the morphological analysis is rational and the average filtering accuracy is better than 90% for different landscapes, especially in the complex landscape with a mixture of buildings and vegetation on the slope terrain.

关 键 词：多尺度 形态学分析 点云滤波 

分 类 号：TN29[电子电信—物理电子学]