Research On 3D Vision Method Based On Binocular Stereo Matching

Vision provides a large amount of rich spatial information.The three-dimensional vision based on the binocular system is regarded as the highlight of the development of computer vision,and this advanced field is developed and integrated by more and more scientific and technological research and development.Using binocular vision technology to analyze environmental information to process the objective world,and quickly realize the scene description in virtual reality from any angle,with the characteristics of low cost,simple process and high accuracy.This article adopts binocular stereo vision method,through camera calibration,stereo matching and 3D reconstruction technology,imitating the human vision principle to restore the real appearance of the actual space target.At the same time,corresponding improvement schemes are designed for the core technology problems.The main work is as follows:(1)According to the principle of camera imaging,the camera model is calibrated by Zhang Zhengyou method.First,This method selects the experimental equipment,constructs a suitable checkerboard calibration object and take images,and obtains the optimal internal parameters of the two cameras through single target calibration;Then the external parameters between the binocular cameras are calculated according to the pose parameter information of the two cameras,and the external parameters are transformed into stereo correction to obtain the parallel binocular system.(2)After fully analyzing the current stereo matching methods,an improved feature-based stereo matching algorithm is conceived.Before matching,the image quality is improved through the pre-processing method of linear contrast broadening,and the matching effect of the discontinuous parallax and the weak texture area is enhanced;The feature matching method is used to obtain basic seed data,and the region growth matching is densified to make up for the lack of feature information.At the same time,it is supplemented by the parallax continuity constraint and the epipolar constraint to improve accuracy and efficiency.According to the calculation characteristics of the matching cost Simplify similarity calculation with integral graph;In addition,in order to make the disparity more precise and dense,the methods of sub-pixel fitting,weighted median filtering,and neighborhood disparity estimation are introduced for refinement processing.After a large number of experiments,the feasibility of the improved algorithm is verified,and a more significant parallax result is obtained.(3)3D visual reconstruction of the entire surface.The disparity data is calculated through the triangulation method to calculate the three-dimensional coordinate value,the color information of each point in the color image is decomposed and integrated into the corresponding point cloud,and the PCL point cloud library is used to establish a four-dimensional space point cloud image with real colors;For erroneous data such as outliers in the point cloud,the radius filter is used as an optimization method for the point cloud to correct,and finally a better 3D visual image is obtained.