密集点云的数字表面模型自动生成方法

Method of Digital Surface Model Automatic Generation with Dense Point Clouds

针对机载航空多视影像密集匹配得到的原始点云存在大量的噪声和空洞,难以直接用于城市数字表面模型重建的问题,提出一种结合增强点云进行法向量优化的泊松表面重建方法。首先通过反投影误差约束和点云距离分布统计分析方法剔除尽可能多的噪声,并通过k邻域均值采样填补点云空缺得到增强点云,及采用固定视点法简化法向量一致化。其次,针对重建表面数据冗余的问题,在保持特征的前提下,引入最短边准则剔除大量的狭长三角形。最后采用ISPRS倾斜下视航空影像及无人机影像进行了实验。结果表明,该方法相比二维狄洛尼构网算法和快速三角化方法在表面重建效果上有一定的改进,对于多视影像的DSM自动生成是可行的。

Due to high level noise and a large number of holes in point clouds generated from multi-view stereo imagery,it is hard to get urban digital surface model directly. A novel strategy combined with preprocessing point clouds and normal optimization for Poisson reconstruction was proposed. Firstly,noise was removed as much as possible according to re-projection error and statistical analysis of distance. Then, resample original point clouds based on average elevation of k neighbor points by k-d tree, and orient all normal vectors consistently toward viewpoint. Finally, to counter the redundancy of reconstructed surface data,a rule of shortest edge in triangle was introduced to restrain bad triangles while maintaining the feature. In experiments with ISPRS oblique nadir aerial images and UAV images, the result proves that, when comparing with the 2D Delaunay reconstruction and fast triangularization,the novel strategy improves the details of reconstruction surface, and can be available for automatic generation of DSM.