面向对象的植被与建筑物重叠区域的点云分类方法

Object-based Point Clouds Classification of the Vegetation and Building Overlapped Area

在分析LiDAR点云数据分类现状的基础上,针对植被与建筑物重叠区域分类困难的问题,提出了一种基于面向对象的点云分类方法。首先采用三角网渐进内插的滤波方法将点云分为地面点和非地面点,并得到DTM;然后对高出DTM一定高度的非地面点建立三角网,删除较长的三角网的边(地物间的边),从而将非地面点云分割成多个对象;再利用各个对象内的三角网坡度信息熵大小判断该对象属于植被或建筑物;最后对于难以区分的对象(植被与建筑物重叠区)根据建筑物几何规则形状延伸扩充,从而提高植被和建筑物重叠区的点云分类准确率。实验结果表明,该方法能够很好地区分建筑物和植被点,分类准确率达到87%。

This paper proposes an object-oriented point clouds classification method for solving the difficult classification problem for the overlapping between vegetation and buildings based on reviewing current status of LiDAR point clouds classification approaches.In the proposed method,the point clouds are firstly separated into ground points and non-ground points through adaptive TIN filter method,and the DTM is obtained.Second,a triangle network is constructed for non-ground points higher than DTM.The non-ground point clouds could be divided into multi-objects by removing longer edges（edge between ground and object）.Then,the object is judged to decide whether it belongs to vegetation or building according to its information entropy of triangle network slope.Finally,for objects difficult to be distinguished from other objects,the overlapped area between vegetation and buildings is extended by geometric shape of buildings,so that the accuracy of point clouds classification of the overlapped area could be improved.The experiment results show good classification performance for buildings and vegetation,and the accuracy reaches 87%.