The Study And Realization Of Multi-View Image-based 3D Reconstruction Algorithm

With the rapid development of virtual reality technology and free viewpoint video technology, three-dimensional reconstruction has become an important cross research area between computer vision and computer graphics. In the light of its low-cost, easy-to-use, etc., three-dimensional reconstruction technique based on multi-viewreceived extensive attention.Against this background, this paper studied multi-view three-dimensional reconstruction algorithms. Multi-view three-dimensional reconstruction, that is, input multi-viewpoint images and camera parameters, through a variety of algorithms, output the three-dimensional reconstruction of the scene target.On the basis of extensive literature research, this paper carried out research to try to complete the following tasks:1. First, the current commonly used camera calibration methods were analysed and researched. There are Tsai camera calibration method, Weng camera calibration method and Zhengyou Zhang camera calibration method. Finally, through the Camera Calibration ToolBox for Matlab, calibration was realized, and results were saved for the three-dimensional reconstruction step.2. Second, the current popular three-dimensional reconstruction algorithms were specifically analysed. There are Visual Hulls algorithm, Space Carving algorithm, and three-dimensional reconstruction algorithm based on Graph Cut.3. The focus is to analyse and realize the following algorithms: EPVH (Exact Polyhedral Visual Hulls), Bundler (Structure from Motion for Unordered Image Collections), PMVS (Patch-Based Multi-View Stereo Software), based on the existing algorithms, improvements were proposed, and experimental simulation results are given, the test speed and three-dimensional effect are both improved.