Abstract: There are a large number of non-ground points in the origin oblique photography point cloud of the construction site. Therefore, it is necessary to adopt appropriate algorithms to filter the non-ground points and extract the ground points. Most of existing filtering algorithms are designed for LiDAR point clouds, and generally have defects like complex parameter settings and poor terrain adaptability. On the basis of cloth simulation filtering (CSF) algorithm, this paper proposes an improved CSF algorithm combining target detection algorithm and terrain analysis method for oblique photography point clouds,implements automatic filtering and ground extraction of oblique photography point clouds in complex scenes.Taking a construction site in Shenzhen as an example, the experimental results show that the type Ⅰ, type Ⅱ error and total error of the improved algorithm proposed in this paper are significantly lower than those of the conventional algorithm, which are 1.11%, 6.36% and 2.39% respectively,and the Kappa coefficient reaches 93.45%,which effectively improves the precision and robustness of the ground extraction of oblique photography point clouds in the construction site.

Key words: oblique photography point cloud, improved cloth simulation filtering algorithm, target detection, terrain analysis, construction site