Research On Improved Algorithm Of 3D Reconstruction Based On Binocular Vision

At present,3D reconstruction technology based on depth from stereo has become study interest of computer vision.The technology has been used in the territories of printed matter detection,geometric measurement,simulate route and digital human medicine.At the same time,this technology also has some shortcomings,such as error,low matching precision of feature points and long running time when calibrating camera internal and external parameters.To solve these problems,In this paper,an adaptive corner detection method and a feature point matching method based on genetic mutation algorithm are improved.The least square method is bringed in the calculation of camera parameters.And regression focal length,pixel center and horizontal rotating movement matrix are also bringed.Through these improvements,more ideal camera parameters are obtained.The matching accuracy is improved.And accurate 3D coordinates are obtained.For the camera parameter error problem,the transformation from object point to pixel point through different coordinate system is analyzed respectively.According to the camera model,the corresponding relationship between object point and pixel point is simplified,the linear equations are determined to solve,and the least square method is used to solve the camera parameters.In determining the object point and the location of the image point value,improve traditional Harris corner detection algorithm,iterate over all pixels of the image and select the appropriate size threshold,adaptive search angular point,and determine the pixel point to the subpixel accuracy,to provide accurate data for subsequent calculations camera parameters,and accurate 3d coordinates numerical calculation.To solve the problem of low matching precision of characteristic dots and improve the matching precision.In this paper,the idea of genetic mutation algorithm is introduced into feature point matching method.The methods combines the inherent constraint attributes between two images and the genetic mutation algorithm to simplify the matching search by the constraint conditions.And the genetic mutation methods calculates the best matching set.The algorithm first determines the position of the pixel matching at the beginning by traversing the image through the window to look for the matching of feature points,calculates the matching set,then finds the best matching point set according to the genetic variation learning algorithm,and finally calculates the depth information and X and Y axis coordinates.In the calculation of three-dimensional coordinates,the principle of parallax is applied to analyze the position difference of the same object in two images,and combining with the parameters of imaging model,the three-dimensional information of images can be accurately recovered.In this paper,a depth from stereo vision experimental terrace was establish to verify the related methods of 3D reconstruction.The 3D reconstruction work is introduced from the aspects of imaging case,camera demarcation,depth information acquisition and 3D depth information solution,relevant methods improvement,experimental results and error analysis and summary.