基于多尺度高程变异系数的影像匹配点云滤波方法

A Filtering Method for Image Matching Point Cloud Based on Multi-scale Elevation Variation Coefficient

针对现有滤波方法在低矮植被密集覆盖区域处理效果差的问题,该文根据不同尺度下无人机影像匹配点云数据所表达的地形地物特征不同,提出基于多尺度高程变异系数的影像匹配点云滤波方法。首先,通过不同尺度的虚拟规则网格构建不同分辨率的DSM,将任意两个不同分辨率的DSM进行差值计算,得到对应两个尺度下的地表特征差异(高程变异程度);然后,对差值DSM计算高程变异系数,根据地物边界区域高程变异系数远大于地形区域的特征进行阈值分割;最后,分析计算高程变异系数的最佳邻域,讨论最佳分割阈值的设定。与传统CSF、TIN和渐进式形态学滤波方法对低矮植被密集覆盖区域的对比实验结果表明,该文方法在低矮植被密集覆盖区域能准确剔除植被点并保留地面点,其中Ⅰ类、Ⅱ类误差分别为9.20%、5.83%,平均总误差为7.68%,均优于CSF、TIN和渐进式形态学滤波方法,可为后期快速建立高精度DTM奠定基础。

Aiming at the problems of over-segmentation,under-segmentation and low accuracy of existing point cloud filtering methods in dense low vegetation areas,a filtering method based on the multi-scale elevation variation coefficient is proposed according to the different terrain features expressed by UAV image matching point cloud data at different scales.Firstly,digital surface models(DSMs)with various resolutions are built using hierarchical virtual grids.Diverse topography elements are captured by digital surface models at different scales.Secondly,the difference of the DSMs(DoD)is obtained by subtraction operation for multi-scale DSM.Thirdly,the elevation variation coefficient(EVC)of DoD is calculated,and the threshold segmentation is performed according to the feature that the elevation variation coefficient of the boundary area is much larger than that of the terrain area.Finally,the optimal neighborhood radius for calculating the EVC and the optimal segmentation threshold for EVC are analyzed.The results show that the proposed method can accurately remove vegetation points and retain ground points in dense low vegetation areas,and the type I error,type II error,and average total error are 9.20%,5.83%,and 7.68%,respectively.The result proves that the proposed method is superior to cloth simulated filtering(CSF),triangulated irregular network(TIN)and progressive morphological filtering algorithm.It can lay a foundation for the rapid construction of high-precision DTM in the later stage.