3D Change Detection of Buildings Based on Multi-level Segmentation of Dense Matching Point Clouds from UAV Images

Abstract

Objectives Buildings are one of the main bodies of the city. Change information of buildings is of great significance to the investigation and treatment of illegal buildings, urban planning management and the real-time incremental updating of geographic databases. Rapid, accurate and low-cost methods of 3D change detection have received more and more attention. Methods This paper proposes a method of building change detection base on dense matching point clouds from unmanned aerial vehicle (UAV) images. First, we meshed point clouds, and analyzed the space features including normalized digital surface model (nDSM) and differential digital surface model (dDSM) and the spectral features including normalized excessive green index (nEGI) and brightness in grids. Then point clouds are segmented based on region growth, and the state of the segmentation object (changing/unchanged/uncertain) is judged. The segmentation criterion will become more and more strict for the segmented object with uncertain state until the changing state of all points is judged(taller/lower/uncertain). Finally, spatial and spectral features are integrated to identify building targets from changed objects. Results The dense matching point clouds from UAV images in two phases of Wuhan University are used to verify the experimental results of this method. The experimental results show that the integrity, accuracy and detection quality of the proposed change detection method can all reach more than 90%. Conclusions The proposed method can achieve object-level, high-precision 3D change detection of buildings based on multi-level segmentation and voting strategy.