Research On 3D Reconstruction Algorithm Based On Binocular Vision

With the rapid development of society,the machine vision technology plays an important role in all walks of life.In machine vision,binocular vision technique is an important branch,and is a direct sign of intelligent machine.This paper is based on the binocular vision system,which carried out the camera calibration,image preprocessing,feature extraction,stereo matching and three-dimensional reconstruction.Firstly,in the stage of camera calibration,for the lack of Zhang's method of extracting corners,the wavelet multi-scale transform is introduced into Harris corner detection algorithm,then the multi-scale Harris operator introduced into the Zhang's calibration method.Secondly,in the image preprocessing and feature point detection stage,there are a lot of noise and difference on the image pair itself.Wavelet noise reduction,median and Gaussian filtering denoising,Histogram equalization to balance the brightness differences,and Wallis sharpening method are used to complete the image preprocessing.For the next stereo matching,use the improved Harris algorithm to detect corner points.Thirdly,for the large amount of computation and complex process in the global stereo matching algorithm.After the scene segmentation is done by OTSU binary segmentation method,the initial parallax is obtained,and then the mismatched points are removed by the left and right consistency.The initial price is corrected by pixel confidence and plane fitting,and the hierarchical belief propagation completes the second matching.Finally,in the three-dimensional reconstruction stage,the method of direct subdivision is used to simplify the space points in the triangle mesh for the reconstruction.Combine the internal and external parameters of the camera to complete the three-dimensional reconstruction.