Dense Scene Reconstruction Of Three Dimensional Space Point Selection Under Double Constraints

With the rapid development of computer technology,both from hardware and software can meet the needs of people better and better.In the development of computer technology,3D reconstruction has become the focus of machine vision,and It has wide applications in cartography,architecture design,construction and construction,homeland security and national defense,entertainment,games and e-commerce,manufacturing,resources and energy,cultural heritage protection,and scientific analysis.The purpose of 3D reconstruction is to reconstruct the 3D model of the scene using real pictures.In the large outdoor scene,the dense algorithm PMVS based on the face has become the hot spot of research for its good performance in all aspects.This thesis makes a detailed study of the original PMVS algorithm and improves it.The PMVS(patch-based multi-view stereo)algorithm is widely used in the multi-view stereo field because of its good performance.However,the surface details of the reconstructed model are lost and the exact location of the reconstruction points is not accurate,these problems are very serious in some cases,especially while the input images are few and the texture of the reconstructed model is less.In order to solve these problems,this thesis studies the removal of mismatched candidate points and the ranking of seed point reliability: firstly,remove candidate mismatch points by using the USAC(Universal Random Sample Consensus)algorithm,and then a double constraint condition strategy is proposed to select the point where the candidate space point has a higher confidence level as the seed point.The details of the reconstruction model are more fit to the original object.Meanwhile the number of reconstruction points in the model with less texture has been significantly increased and the holes of the texture model are obviously reduced.The validity of the improved algorithm is proved to be more effective and practical.Through experiments on real data sets,the error rate of point cloud generates in dense point cloud is increased by 0.04~6.905,which makes the detail problem of point cloud extension well solved.The improvement rate of point cloud patches is 3.873~34.612,so that the problem that the texture is not too obvious to reconstruct less point clouds has been solved.The agreement between the reconstruction model and the original object has been greatly improved,and the number of reconstruction points in the less texture model increases obviously,and the vulnerability is obviously reduced.The experimental data verify that the improved algorithm is more effective and practical.This thesis uses the campus of Harbin University of Science and Technology as an example for building 3D reconstruction and uses the 3D reconstruction model to build a visual management system.The system has statistical personnel information function and personnel information query function through related fields.At the inquiry stage,the system connects the local server firstly.The server will parse the incoming database query statements and then extracts the corresponding data from the database,after the data extraction stage,the corresponding data information is displayed.Thus,the 3D reconstruction technology is applied to the visualization management.