Global Parameter Estimation In 3D Point Cloud Reconstruction

3D reconstruction is a typical visual problem in the field of computer vision,especially worthy of attention is 3D reconstruction technology based on image.Because of its low cost?low precision and high flexibility and other characteristics,it is attracted many scholars involved in the research work and widely used in many fields.With the deepening of the research,some of the technical problems have been solved well.However,the estimation of the motion parameters has always been a difficult point of study and has not been well solved.At present,the research mainly focuses on the calculation of the camera motion parameters,the purpose of the estimation is to obtain more accurate camera parameters,which can be used to reduce the sensitivity to errors and obtain high precision three 3D point cloud.In the end,this paper does the following research and work on the basis of multi-view geometric reconstruction.(1)Based on the existing feature matching algorithms,the classical fixed threshold method is redesigned,and an adaptive threshold method is designed to determine the threshold size according to the specific conditions of different data sets to obtain the initial roughness Matching point pairs.At the same time,epipolar geometry constraints in multiview geometry are introduced to replace the traditional homography matrix constraints,and the fundamental matrix is estimated by RANSAC algorithm for elimination of error matching points.The experiments on three different data sets show that the improved matching algorithm can indeed improve the matching accuracy.(2)For the estimation of motion parameters,this paper adopts the idea of global optimization and considers all cameras in the same world coordinate system for optimization.Firstly,obtaining the matching point of the input image,the relative two-point rotation matrix is calculated by using the matched point pair,and then all the absolute rotation matrices are optimized according to the pairwise relative rotation.Then,according to the principle of epipolar constraint in multi-view geometry and the matching point pairs,a linear system of equations is constructed,and finally all the absolute translation vectors are obtained linearly and linearly.(3)Programmed to achieve improved image matching module,the camera motion parameter recovery module,the space structure recovery module,combined with other preprocessing modules,dense matching module,a complete three-dimensional point cloud reconstruction project is built finally,the results on three different data set show that this method can get a good point cloud and improve the accuracy of camera motion estimation.