Research On Three-dimensional Reconstruction Of Binocular Vision Under Uneven Illumination

In the field of computer vision,three-dimensional reconstruction refers to the process of using computer algorithms to reconstruct the threedimensional shape information of an object from a single or multiple view image.The binocular vision system is one of the important ways to achieve three-dimensional reconstruction by simulating human eyes to obtain three-dimensional information of spatial objects.The basic information of3 D reconstruction based on binocular vision is the image data collected by the camera.In practical applications,due to complex lighting environments or surface reflection of objects,the collected images often exhibit uneven lighting,resulting in low contrast,color distortion,and loss of details,thereby affecting the effectiveness and quality of 3D reconstruction of objects or scenes.This article mainly studies the three-dimensional reconstruction method of binocular vision under uneven lighting conditions,with a focus on improving non-uniform image enhancement,binocular camera calibration,and stereo matching algorithms.The specific work is as follows:(1)In order to meet the performance requirements of non-uniform illumination images,a non-uniform image enhancement scheme is designed.Firstly,a theoretical analysis of the specific performance of nonuniform images is carried out,and then based on the principle of color constancy and the theory of frequency filtering methods,an image enhancement method under non-uniform illumination conditions based on Retinex is proposed.Experiments using commonly used quantitative evaluation methods show that the algorithm proposed in this paper is effective and enhances the details of uneven images.(2)In order to improve the accuracy of camera calibration results,the camera calibration method has been improved.Firstly,the binocular vision system model and the transformation relationship between spatial point coordinates and pixel coordinates during 3D reconstruction are analyzed.Secondly,aiming at the fact that the accuracy of corner extraction from the chessboard calibration board in Zhang Zhengyou’s calibration method is easily affected by noise and image quality,a binocular camera calibration method based on dot array is proposed.Finally,controlling other variables to remain unchanged,a comparative experiment was conducted using a checkerboard calibration board and a dot array calibration board,respectively.By comparing the camera parameters obtained from the comparative experiment with the actual camera parameters,the feasibility of the algorithm in this paper was verified,and the accuracy of camera calibration was improved.(3)In order to improve the clarity and noise resistance of disparity images obtained by stereo matching,an improved AD-Cenus stereo matching algorithm was designed.Firstly,the general steps of stereo matching are briefly introduced.Secondly,in the cost calculation stage,a stereo matching algorithm based on improved Census transform and AD value fusion is proposed to solve the problems of Census transform that is too dependent on the gray value of central pixel points and has poor antiinterference ability.Then,a clearer disparity map is obtained by optimizing the cost aggregation of the "cross" region using scan lines,calculating the "Winner Take All"(WTA)disparity,and optimizing the left and right consistency test.Finally,through comparative experiments of three algorithms,it is verified that this strategy retains the robustness of the Census algorithm and reduces the mismatch rate.(4)In the 3D reconstruction section,the depth map containing the pixel depth values of each point position in the actual object image obtained through stereo matching is converted into point cloud information through calculation.The Open3 D open source library is used to process the point cloud data and reconstruct the 3D information of the actual object.The experimental results show good reconstruction results.