A Research Of 3D Reconstruction Technology Based On Multiple View Geometry

The 3d reconstruction technology based on multiple view geometry is an important topic of computer vision,which has great values in applications on location and navigation,virtual reality and argumentation reality,medical imaging assisted diagnosis and therapy,etc.The Thesis argues the most common 3d reconstruction technology based on a monocular camera: evaluate 3d positions of the scenes and the objects in the space by taking many 2d photos,then finally restores the 3d models of the objects and the scenes.The 3d reconstruction technology consists of three parts,structure-from-motion,multi-view stereo,surface reconstruction.In the experiment,we capture dozens of photos by a monocular camera surrounding the real object,and by using these photos,finally generate the 3d model with high realities.In part of the structure-from-motion,the thesis computes and locates positions and poses of the cameras and 3d coordinates of the objects by the theorem of multiple view geometry,optimizes the result of the structure-from-motion by using bundle adjustment,and proposes a fast algorithm based on conjugate gradient algorithm to improve the resolve of the bundle adjustment algorithm which uses GPU accelerating technology,shortens the time to one fifths approximately.In part of the surface reconstruction,the thesis uses Poisson reconstruction algorithm to reconstruct 3d models of the scenes and the objects by generating 3d meshes from sparse point clouds,finally post-processes the 3d model by digital geometry algorithms such as mesh simplification and mesh smoothing to optimize the result of the 3d reconstruction,proposes an adaptive mesh simplification algorithm based on k-means clustering which can preserve the reconstruction quality under the condition that decimating faces to one tenth as well after the simplification.In conclusion,the thesis optimizes the structure-from-motion and the surface reconstruction in the 3d reconstruction,and promotes the efficiency of the reconstruction on the condition of preserving qualities.