3D Change Detection Of Buildings Using Dense Matching Point Clouds From UAV Images

Buildings are one of the main bodies of the city.Building change information is of great significance to the investigation and treatment of illegal buildings,urban planning management and real-time incremental updating of geographic databases.Building change detection methods can be divided into two-dimensional change detection and three-dimensional change detection.The two-dimensional change detection methods mainly use satellite remote sensing image and aerospace image as the data source.The change detection results are susceptible to the non-linear change of gray level and perspective deformation caused by different viewpoints in different periods,as well as the influence of shadow and projection difference.The threedimensional change detection methods mainly use airborne LiDAR(Light Detection and Ranging)point cloud as the data source,but the LIDAR point cloud is expensive to gather.Traditional building change detection methods have obvious shortcomings and limitations in terms of cost,efficiency,and accuracy.Rapid,accurate and low-cost three-dimensional change detection methods have received more and more attention.With the development of UAV(Unmanned Aerial Vehicle)hardware and MVS(Multiview Stereo)technology,dense matching point cloud from UAV images provides a lowcost data source for building change detection.However,due to the limitations of the quality of the point cloud,the occlusion of data and the lack of change sample base,the existing change detection methods are difficult to be directly applied to the change detection using dense matching point clouds from UAV images.In this paper,a building change detection method based on multi-level segmentation is proposed based on the geometric and spectral information contained in dense matching point clouds from UAV images.In this method,the dense matching point clouds of multi-temporal UAV images have meshed firstly,and the nDSM(normalized Digital Surface Model)and dDSM(differential Digital Surface Model)geometry features and nEGI(normalized Excessive Green Index)and brightness spectrum features within the grid are calculated;Then,point cloud segmentation based on region growth is carried out,and the change / unchanged / uncertain state of the segmented object is judged.For the segmented object with uncertain state,step by step strict growth criterion is employed to iterative segmentation,until the changing state of all points is judged.Finally,geometric shape features and spectral features are integrated to identify the three-dimensional building targets in the changing objects.In this paper,two phases of the dense matching point cloud from UAV images of Wuhan University are used for experimental verification.The results show that the threedimensional change detection method proposed in this paper is more than 90% in the completeness,accuracy,and quality of building change detection,and has a sharp calculation efficiency and detection accuracy.The object-level 3D building change detection using dense matching point clouds from UAV images is realized.