Research On Building Modeling And3D Change Detection Technology

With the rapid development of remote sensing technology, the stereo imagery ofLiDAR data obtained from satellite platform of aviation platform can be used to extractaccurate3D information. It makes a great positive influence to3D reconstruction forbuilding models and makes the digital city to become a reality. In recent years, ourcountry accompanies with frequent natural disasters. Due to this, the problem aboutdisaster assessment and disaster management appears especially important and3Dchange detection is the key technology to solve this problem. This thesis focuses on twoaspects, model building and3D change detection, aiming to promote the digital cityconstruction and disaster assessment.Firstly, this thesis introduces the method of3D information extraction for a helpfuldata support. One one hand, this thesis dicusses3D information extraction based onstereo imagery. The mathematic description of optical sensors is given first, whichincludes strict sensor model and gerenal sensor model. It explains the algorithm offorward intersection and resection. On the other hand, this thesis introduces the basictheory about LiDAR system and the method about generating DSM data from originalLiDAR data, as a helpful support for subsequent application.Secondly, this thesis provides a novel level set framework for3D building modelsin LOD2with geometry structure of typical roofs. Local information is introducedtowards multiphase and multichannel level set function. The energy function isminimized when each part of roof data corresponds to the same normal vector as featurevalues for level set segmentation. The advantage of this method is that for complexbuilding models, roof primitives as well as roof topology graph can be extracted fromhigh-resolution DSM data with high accuracy, evaluated by completeness ofsegmentation and RMSE of3D reconstruction. The very promising experimental resultsdemonstrate the potentials of our method for large-scale building reconstruction inLOD2.Finally, with3D change detection as main application, according to LOD model,this thesis puts forward CDL (Change Detection Levels) model, which is used forquantitative assessment of3D change detection in different levels. In order to realize3Dregistration of3D data in different times, this thesis uses the ICP algorithm that canquickly and effectively solve the problem of dislocation of data. After that,3D changedetection uses adaptive threshold to adapt to different datasets. Meanwhile, it givesreasonable and systematic quantitative evalution, to meet different application.