Research On3D Building Model Reconstruction From Airborne Lidar Data Based On Contour Clusters

The high-precision DEM extraction and3D building model reconstruction has been a hot research field in seurveying, mapping and remote sensing. Currently, bhotogrammetry technology is still the most economical and efficient means for the cconstruction of the wide range of3D terrain and large scenes of CyberCity, but still lower level of process automation and thus a lot of manual intervention and editing is required to complete the mission. In recent years, the rapid development of the LiDAR (Airboren Light Detection and Ranging, LiDAR) technology provides a new solution for rapidly acquiring the wide range of terrain and3D city model. Point cloud filtering is one of the technologies which have been concerned earlier by many scholars in LiDAR field. Despite a lot of point filtering algorithms have been proposed and achieved certain results, but these methods largely focus on a particular terrain category, universality is not enough. Moreover, filtering results in complex environment is still not ideal. In addition, the detection of the building, in a high degree of vegetation cover, especially in the case of buildings and trees adjacent effectively, extracting the outline of buildings is also an important issue. At the same time, as regard to the3D reconstruction of the building model, existing programs generally focus on the reconstruction of the roofs composed by plane facets. The relevant discourse about the reconstruction of the curved edges and curved roof is still rare. Therefore, this paper carries out some deep research on LiDAR point cloud data filtering and3D building model reconstruction.In this thesis, the point cloud filtering technical and the process of building detection in complex environments and the key technologies for automatic building model reconstruction are investigated. As to the building model reconstruction, we focus on how to using contours to retrieval the models.The major works are listed as follows:1) LiDAR point cloud filtering algorithm based on multi-scale analysis. First, a multi-scale grid is built, and then the threshold parameter is adaptively determined by local statistical analysis and taking into account the multi-scale, topographic description error. The ground points are gradually selected from top level to the bottom level and the DEM become more precise stepwise.2) Contour tree contruction and contours clustering. Spatial topological relations of the contours and the manifestations of contour tree have been studied, and then a technical based on parent node tracking for contour tree construction has been proposed. Rough contour clustering can be achieved based on the contour tree pruning operation and the exact clustering can be achieved through contour matching which based on the corresponding point set matching for shape similarity measure. Contours’clustering is the basis of the subsequent process for building3D reconstruction.3) Building detection in the complex environments. A progressive building detection method based on contour cluster features and traditional features has been proposed. Preliminary detection of the building can be done by the using the features as elevation difference between first and last return echo, nDSM elevation, roughness, morphological characteristics, then the multi-return echo point density and shape characteristic parameters of the clusters is calculated for the exactly detecting of building regions. With this method the problem of the adjacent between buildings and trees can be largely solved.4) Research on the reconstruction of3D building model. A3D reconstruction frame is proposed Based on the contour cluster. The segmentation of the multi-level structure building is finished by the contours clustering. The difference between the contour clusters characteristics in2D plane for different types of building model is studied; a curve reconstruction method based on "polarization angle index" is proposed to solve the zigzag noise. On this foundation, the model type of hierarchical structure can be recognized by combining the contour cluster characteristics and curve reconstruction results. According to the model type recognition results, flexible reconstruction methods are selected to construct the hierarchical model. Finally, according to some certain merging rules, the whole building model is completed with the integration of hierarchical models. Experimental results certificated the validity of the proposed method.