Research And Implementation Of Point Cloud Data 3D Reconstruction And Tracking Registration Technology

This dissertation focuses on the point cloud 3D-reconstruction and tracking registration technology,including sparse and dense point cloud three-dimensional reconstruction,accuracy measurement and tracking registration technology.For tracking registration technology,the existing algorithms have poor recognition effect in recognizing scenes with large disturbances such as skewness,scaling,blurring and so on.This paper studies and improves the point cloud three-dimensional reconstruction algorithm and tracking registration technology.For affine-invariant matching,the affine invariant feature points are extracted and matched by combining the maximum and minimum tree and the limit feature region.Aiming at the problem of accuracy measurement of reconstruction results,a new accuracy measurement algorithm is proposed to calculate the accuracy of reconstruction by transplanting the existing point cloud reconstruction process into the virtual scene,and then comparing the original model in the scene with the reconstructed model.For the tracking registration algorithm,the existing methods are improved,and a full-scale high robustness method is proposed to improve the application effect of tracking registration technology,and the point cloud three-dimensional reconstruction model is projected to the real world to realize the practical value of tracking registration technology.In this paper,through a large number of experiments and data analysis,combined with open data sets and original data sets,the effectiveness and advantages of the algorithm are verified.The main contents of this paper including:(1)Proposed the Precise Extreme Feature Region(PEFR)algorithm.Use Morse theory to separate the max-tree and min-tree from the result of extreme feature region algorithm to solve the problem of feature matching for highly affine transformed images and improve the reconstruction accuracy of point cloud three-dimensional reconstruction algorithm.(2)Realize the 3D point-cloud reconstruction method in virtual environment according to research on the virtual environment imaging theory and put forwards the average descending rate(ADR)algorithm to measure the accuracy of point cloud three-dimensional reconstruction results,which realizes the low-cost measurement of point cloud three-dimensional reconstruction results.(3)Proposed the All-Scale Fast Tracking(ASFT)method to improve the recognition ability and matching efficiency on tracking registration and combine the tracking registration technology with the point cloud three-dimensional reconstruction technology to displaying the model generated by the three-dimensional reconstruction method,which realize value of registration tracking method.