Research On Feature Matching Model Of Single-scale 3D Reconstruction Based On Stereo Vision

Since the concept of 3D reconstruction was first proposed,the concept of 3D reconstruction has developed rapidly and expanded rapidly in a short period of more than a decade,and its application scene is also increasing.In common algorithms such as SIFT(scale invariant feature transform),SURF(Speeded-Up Robust Features),scale invariance is realized by using scale pyramid,but it is difficult to meet the real time requirement when the speed is slow.The SURF,SIFT algorithm is no longer suitable for the requirement of real time,and then the ORB(Oriented FAST and Rotated BRIEF)algorithm emerges as the times require,but the ORB algorithm is too simple and often has a large error.At present,there are no algorithms to balance the time consumption and performance effect in the application scenarios with little or no scale change.Therefore,in this paper,the 3D reconstruction technology of this kind of problem is studied in depth,and a 3D reconstruction technique suitable for the single-scale scene is proposed.Firstly,an improved Harris corner detection algorithm based on preprocessing is proposed.The proposed method solves the problem that Harris corner detection has a large amount of computation and time consumption.By comparing the gray values of neighborhood pixels and center pixels,the pixel points whose gray difference is greater than the adaptive threshold passes through the initial screening.Then calculate the Harris response,the point that satisfies the filter condition is corner point.Secondly,the efficiency of RANSAC algorithm is easily affected by the size of subset,the ratio of intra-class points and the size of data set in eliminating mismatch points.In this paper,an improved GMS-RANSAC algorithm based on multi-strategy selection is proposed.Firstly,select the strategy according to the number of feature points extracted,and select right points by GMS.Then the selected feature points are sorted according to the score,and the RANSAC algorithm selects the feature points according to the score level.Finally,the reprojection error score is calculated according to the idea of optimization,and the model with better result is selected adaptively.Finally,in order to solve the problem that the traditional SURF algorithm is slow by constructing the scale pyramid,and the ORB algorithm is fast but ineffective,a new model HGR model of feature point selection and matching is proposed.Improved Harris feature detection,GMS-RANSAC mismatch elimination and ICE-BA optimization algorithm are used for 3D reconstruction.The proposed method is suitable for the case where there is no scale change.The proposed method solves the limitations and shortcomings of the original method.