Research On 3D Reconstruction Technology Based On Binocular Vision

At present,binocular vision 3D reconstruction technology has become an important research direction of computer vision.This technology has been applied in the fields of intelligent automatic production,robot navigation,medical assistance and so on.However,this technology still has some defects.For example,when calculating the internal and external parameters of the camera,the distortion of the lens is easy to cause errors,resulting in inaccurate internal and external parameters.When the feature point detection is sparse,the visual effect of the three-dimensional display is poor;the feature point matching time Long wait questions.In view of the above problems,this paper will improve this based on the three-dimensional reconstruction of binocular vision.Based on Zhang Zhengyou's calibration method,the camera calibration is improved,the lens distortion and the real-time performance of the whole calibration process are fully considered,and the radial distortion which has great influence on the calibration parameters is corrected,and the tangential distortion with little influence is neglected.Not only improves the parameter accuracy,but also the efficiency is high.Based on the SURF matching algorithm,the matching method is improved,and the retrieval efficiency of feature points is improved.The following are the main contents of this paper:(1)For the calculation of internal and external parameters of the camera,the conversion relationship between each coordinate system is analyzed.The Newton iteration method is used to calibrate the radial distortion of the lens.The rotation matrix R and the translation vector T are obtained,and the pre-calibration and calibration are compared.The back corner detection error shows that the calibration accuracy is greatly improved.Then,considering that the scene is affected by various types of noise,Gamma correction and random noise preprocessing are performed separately.The Gamma transform significantly improves image contrast.The nonlinear filtering of random noise is performed by means of median filtering,which can filter out random noise points and retain most of the original information.(2)For the defects in feature detection and matching algorithms,corresponding improvement methods are adopted to improve some shortcomings in the algorithm.Firstly,the sparse feature point detection will lead to poor 3D reconstruction based on feature points.For this reason,this paper combines the triangulation algorithm to analyze and improve it,and the 3D reconstruction has been significantly improved.Secondly,the matching time is related to the advantages and disadvantages of the whole system.The SURF matching algorithm is improved,and the polar constraint idea is integrated into the matching link.Experiments show that the improved matching algorithm not only improves the matching success rate,but also improves the matching success rate.The matching retrieval time is greatly shortened.In this paper,a three-dimensional reconstruction platform based on dual-purpose mineral materials is built,and the three-dimensional reconstruction research is based on this.The software platform design was completed,mainly relying on VS2015 and OpenCV software development platform,respectively,completed camera calibration,feature point detection and matching,3D parallax display and other experiments,and finally integrated sub-modules into a system based on MFC.