Building modeling and monitoring using social media images

Abstract

For urban monitoring and planning, 3D city model always provides a valuable data source. Usually, satellite and aerial images are used to build the 3D city model. Over the last decades, the rapid development of social media has accumulated vast amount of online images which provide a valuable image time series of the city. Therefore, 3D reconstruction using social media images has gained much interest. However, one single static 3D model is not sufficient to fully explore the development of urban environment and to monitor the change of city. Thus a time-varying 3D model should be reconstructed. However, in residential area where social media images are not sufficient for 3D reconstruction, the geometric change can also be detected by comparing images taken at different times for urban monitoring. This thesis is the first attempt to obtain spatial-temporal state of environment using only social media images for urban area monitoring. Keeping the objective in mind, a few algorithms have been developed for either obtaining 3D structure of the scene at different times, or detecting changes. In this thesis, an efficient method is proposed to only build one point cloud and determine the 3D structure of the scene at different times by comparing the 3D point cloud against images observing it. Moreover, this 3D point cloud existence determination method is extended to further deal with the additional challenges introduced with dramatic geometric change. In addition, another method was proposed to detect the geometric change of city without reconstruction by comparing the similarity of image patch. The experiment result of 3D point cloud existence determination method demonstrates its capability of recover coarse geometry of 3D structure of the scene at different times for social media images. Besides, the result of the method that only based on images, named 2D image change detection method, can highlight the area with geometric change.