Researh On Urban Reconstruction With 3D Line Segment Constrain

Recently, With the development 3D reconstruction technology, urban reconstruction is becoming a hot topic of academic research and commercial interest. Image based urban reconstruction technique can reconstruct large-scale urban 3D model with low-cost and high-efficient, This technique use image sequences to generate MVS point cloud, then the MVS point cloud is used to generate surface mesh model. However, Holes, oulier and noise points often present in MVS point cloud. Due to these noisy points, edges of surface mesh model are over-smoothed. Urban 3D model can’t represent sharp building edges and visual effects need to be improved.Aiming this problem, this thesis proposes an urban reconstruction method based on 3D line segment constrain, in order to improve visual effects of urban 3D model and make the surface mesh can represent sharp building edges.The main work is summarized as follows:1. In order to reconstruct 3D line segments model, an algorithm based on MVS point cloud re-projection is proposed. In line segments matching stage, this algorithm uses MVS point cloud re-projection points to match multi-view 2D line segments accurately. In 3D line segments generation stage, this algorithm reconstruct 3D line segments by intersecting 3D line and 3D plane. Then outlier 3D line segments are removed by 3D line segment support counter. Finally, this algorithm use non-linear optimization method to minimize 3D line segments re-projection error and improve accuracy of 3D line segment model.2. A surface mesh refinement algorithm based on 3D line segment constraint is presented in this paper. This algorithm use poisson surface reconstruction algorithm to generate surface mesh model, then it adjusts vertices near 3D line segments to refine the surface mesh model. The experiment shows that the refined surface mesh can represent sharp building edges and improve visual effects of urban 3D model.Through study of above algorithms, a prototype system based on above algorithms is implemented. Some experiments are conducted as well to compare our algorithms with other recent algorithms. The result of these experiments indicate the effectiveness and correctness of our algorithms.