Research On 3D Reconstruction Method Based On Binocular Stereo Vision

The real world are three-dimensional images,and the images obtained through cameras or other devices display two-dimensional plane images and will inevitably lose some useful information such as depth.The research of binocular stereo vision technology is to simulate human eyes through the use of two cameras with different angles to achieve the purpose of perceiving scene depth information and facilitating the reconstruction of three-dimensional objects.This thesis designs a 3D reconstruction system based on binocular stereo vision under the HDevelop development environment of HALCON software.The main works are:1.In the calibration part of the cameras,the cameras are calibrated on the basis of Zhang Zhengyou calibration method,and the distortion parameters are corrected through the polar line constraint to obtain the corrected camera parameters.2.To process the images acquired.Firstly,the obtained color image is transformed into a gray image by the weighted average method.Then,the grayscale image is divided by threshold to extract the region of interest.Finally,improving the contrast of the image through the process histogram equalization.3.Employing an intensive stereo matching algorithm based on multiple grids.The method can reduce the searching range of the matching by using the polar line constraint and the distance between the camera and the target object to estimate the search range of the parallax.The coarse grid series is introduced by the multi-grid technology to accelerate the convergence of partial differential equations to improve the matching speed.The similarity judgment criterion combining gray level,gradient,and smoothness is adopted to avoid over-dependence on the gray value,and it is possible to estimate the disparity value with no corresponding point by smoothness,and the resulting disparity field is smooth and continuous.4.The smoothing filtering of the initial 3D point cloud based on the moving least squares method.Firstly,an approximation surface is provided to the discrete point cloud by a two-dimensional surface fitting function,and then the points distributed around theapproximation surface are moved to the surface through iterative projection to eliminate the point cloud noise and obtain a smooth point cloud surface.5.Using the greedy algorithm to triangulate the smoothed 3D point cloud.First,search for the neighboring points that meet the criteria by the radius neighborhood search algorithm,and project the neighborhood points onto the tangent plane,then use the greedy algorithm to do the local 2D triangulation,and then map the 2D triangulation points back to the space to form Space triangle grid,and the morphological processing is used to fill the black hole on the surface of the grid.The experimental results show that the depth information can be effectively recovered by the multi-grid intensive stereo matching,and the smoothness of the 3D point cloud can be achieved by moving the least squares method.The good visual effects of the three-dimensional reconstruction also indicate the feasibility and effectiveness of the matching method and the deep post-processing method applied in the article.