3D Point Cloud Reconstruction Using Global Linear Method

Three-dimensional point cloud reconstruction using global linear method is a method to recover scene structure information using images.The main research work of this paper includes the following points: feature extraction and matching,calculation of camera motion parameters,triangulation and bundler adjustment.Our contribution is to consider all the cameras in a unified coordinate system,and calculate the motion parameters of the camera through the global structure from motion method to complete the 3D point cloud reconstruction.The main research work of this paper includes:Reconstruction model of camera triplet: In the reconstruction model,this paper uses the coordinates of three cameras and the constraint relationship of a scene point to calculate the proportional relationship of the baseline length.Compared to traditional two-view based stereo vision reconstruction,the advantage of three camera reconstruction is that it contains more information than the two views,avoiding degradation due to the collinear motion of the camera.At the same time,in the feature matching stage,this paper uses feature tracking instead of feature matching to calculate the camera's pose more accurately.Outliers filtering: As there may be many mismatches in the feature matching phase,a lot of abnormal data will be generated.In order to filter out these outliers,this paper designs outlier filtering steps,including constraints in the feature matching process to reduce false matches,introduces loop consistency constraints in the camera triples to reduce errors of epipolar geometry due to repetitive scenarios,and use point transfer constraints in the calculation of translational scaling to reduce the impact of outliers.Multi-view reconstruction: For the traditional incremental reconstruction method,due to the frequent BA,it can easily cause the scence drift error and low time efficiency,we use the method of solving all the camera poses uniformly,so that the cumulative error is evenly distributed,avoiding scene drift and the time efficiency is greatly improved.In the research work of this article,the linear equation is used to calculate the direction and position of the camera in the global coordinate system.The triangulation is used to recover the scene point,and the reconstruction result in multi-view is completed.Finally,the reconstructed result is optimized to obtain a complete and accurate 3D reconstruction result.