Research On Feature Point Matching Algorithm In Binocular 3D Reconstruction

With the development and improvement of computer vision technology,many new technologies have emerged continuously,and one of the most important technologies,the 3D reconstruction technology,has also been widely used.This technology has an irreplaceable position in the future human-computer interaction process.At present,3D reconstruction technology has been widely used in ancient building restoration,medicine,manufacturing modeling,VR(Virtual Reality),AR(Augmented Reality),etc Many fields.However,the traditional 3D reconstruction technology requires the use of additional auxiliary equipment,resulting in a very high hardware cost of the traditional 3D reconstruction method.This fatal shortcoming also limits consumer-level applications of 3D reconstruction technology.How to realize 3D reconstruction with low hardware cost and high reconstruction accuracy has also become a research hotspot and difficulty.The purpose of this article is to use two low-cost network cameras as image acquisition terminals,to improve the accuracy of feature point matching in the 3D reconstruction process,and to improve the 3D reconstruction effect without increasing hardware costs.The specific content includes:First,the binocular method is analyzed and determined to realize low-cost 3D reconstruction.Based on this,the algorithms involved in the binocular 3D reconstruction process are further introduced.At the same time,based on the experimental environment,the required hardware selection and the development environment and specific software design work needed to implement the software part of the 3D reconstruction system are introduced.Secondly,the experimental simulation is based on the traditional 3D reconstruction algorithm.Select a representative 3D reconstruction scene as an example picture and follow the binocular image acquisition,camera calibration and image correction,FAST(Features From Accelerated Segment Test)feature point detection and SIFT(Scale-invariant feature transform)feature point description,The steps of feature point matching,world coordinate system calculation,triangle surface mesh reconstruction,and Open GL(Open Graphics Library)reconstruction mapping were performed on the sample image based on the traditional3 D reconstruction algorithm.Thirdly,an improved algorithm is proposed for the problems of mismatching feature points and 3D model distortion in the binocular 3D reconstruction algorithm.The algorithm is improved from three aspects: feature point filtering based on the matching result score,feature point filtering based on perspective transformation,and feature point filtering based on the image subject.The optimization of these measures can effectively filter out the mismatched feature points during binocular 3D reconstruction,thereby improving the accuracy of 3D reconstruction.At the same time,a binocular 3D reconstruction system is built based on the improved algorithm.The improved algorithm is compared with the unimproved algorithm in terms of eliminating feature point mismatches and improving the effect of 3D reconstruction models,and the results are compared with the traditional 3D reconstruction software Visual SFM(A Visual Structure from Motion System)and based on Quantitative comparison of infrared RGBD(RGB and Depth)feature point clouds.Finally,several experiments were performed in different scenarios.The experimental results show that after the algorithm is improved,most of the mismatched feature points can be effectively eliminated in different scenarios.The effect of the improved 3D reconstruction algorithm is much better than the traditional 3D reconstruction algorithm.