Research And Implementation Of 3D Point Cloud Reconstruction Based On WEB

In computer vision and computer graphics,3D reconstruction is a common method for obtaining the shape and appearance of a real object.At present,there have been in-depth research and development in different directions.The three-dimensional reconstruction scene model reflects richer and more intuitive information than the two-dimensional plane,and is widely used in digital cultural heritage protection,game movie scenes,medical images,three-dimensional navigation,virtual reality and other industries.Based on the stereoscopic vision reconstruction method,this paper realizes the reconstruction and display of the object point cloud model from the WEB end.The main research contents of the paper include camera calibration and sparse point cloud generation,multi-view stereoscopic reconstruction,WEB-side automated dense point cloud reconstruction,and point cloud meshing.The main work of the thesis is divided into the following sections:The first part is camera calibration and sparse point cloud generation.This paper uses a self-calibration method from the motion recovery structure.Using an ordinary camera,multiple images are taken from multiple perspectives on objects that need to be reconstructed,the pose of the camera is calculated by matching feature points between pairs of images,and camera internal parameters are obtained from the camera's EXIF information,combining the previous two camera gestures and interiors.Parameters,triangulation to restore feature points in the three-dimensional space,the establishment of sparse point cloud.The second part is multi-view stereo reconstruction.In this paper,a patch-based dense point cloud reconstruction method is used.Combined with the camera parameters and the sparse point cloud generated in the first part,the dense point cloud of the model is generated through the steps of feature point matching,expansion and matching filtering.The third part is to realize an automatic reconstruction system based on WEB.Based on the combination of the first two parts,the WEB-side automated dense point cloud reconstruction is realized.It is only necessary to upload a plurality of images from different perspectives of the object on the website,and after the submission,a dense point cloud model of the photographed object can be automatically rebuilt,and the object point cloud model will be displayed in the browser after the reconstruction is completed.The fourth part is the study of the grid of point clouds.Using the rolling ball algorithm and Poisson surface reconstruction algorithm respectively,the point cloud model is meshed,and the time and mesh integrity indexes of the mesh models generated by the two point cloud meshing algorithms are compared and analyzed.Exploring the application scenarios of the two algorithms.