Research For Some Key Technologies On 3-D Reconstruction Of Scenes

With development of computer technology and demand of reality, 3D reconstruction of scenes have become a hot topic. In this dissertation some key technologies of 3D reconstruction was studied, and corresponding scheme was proposed.Firstly, the development and present situation of 3D reconstruction were reviewed, and a few classic methods of 3D reconstruction were discussed. Feature matching of the same scene was a key problem in reconstruction based on image. In order to improve the performance of present feature matching, an algorithm of Feature Matching Based on Corner Affine Invariant was proposed. It selected corners as features, removed influence of image stretch, skew, rotation, and translation by calculating affine invariant, so matching accuracy and speed were improved.Accurate reconstruction of free-formed objects was very difficult, owing to error of feature extracting and matching in 3D Reconstruction. A method that reconstructs object by using Quadric Spline Curves as primitives was proposed. It can reduce the effect of approximate error, and increase fitting accuracy in curve fitting.Finally, genetic algorithm based on the Earliest Finish Time heuristic was proposed to improve rendering speed of scenes. Because of intrinsic parallelism of genetic algorithm, it can be applied to rendering of distributed scene. Priorities of task instead of problem solution were used in Chromosome encoding. EFT was used in search heuristic, which attained fine convergence speed and shortest scheduling length. The algorithm had good performance in distributed scene rendering.