关键词: 点云分类, 建筑物变化检测, 布料模拟滤波算法, 动态图神经网络, 三维变化

Abstract: Detecting building changes is an important task to analyze the change of urban spatial layout. To address the problem of noise or complicated boundary in the process of change detection using satellite images,this paper researches on the automatic extraction of change information of buildings for plane and height from the point cloud reconstructed by photogrammetry techniques. Firstly, the cloth simulation filtering algorithm is proposed to reduce the influence of terrain, and then a deep learning technology of dynamic graph neural network is used to effectively extract the building points. The change regions of buildings are extracted through the comparison of two point cloud classification results. This paper selects the dense matching point clouds of two phases of tilt photogrammetry in some areas of Xiaoshan District, Hangzhou for experimental analysis and verification.The results demonstrate that the proposed method can quickly realize reliable building change detection in a wide range, the change information of plane and elevation dimensions of buildings is well reflected,and can provide support for fine urban management.

Key words: point cloud classification, building changing detection, cloth simulation filtering algorithm, dynamic graph neural network, 3D changes