Research On Feature Matching Algorithm In 3D Reconstruction Of Binocular Vision

Vision is an important way for humans to acquire information and to recognize the world.With the continuous improvement and improvement of related disciplines such as image processing,electronics,and optics,computer vision technology has become more and more mature in human observation and brain processing information,and has been used in military,medical,agricultural,aerospace,machinery manufacturing,and forestry operations.There are more and more applications in other fields.Nevertheless,computer vision still cannot be as flexible and versatile as human vision.Many factors in the reconstruction process will affect the accuracy of reconstruction.Therefore,research on computer vision technology still has important research value.Binocular vision simulates the human binocular vision system,uses stereoscopic image matching technology to capture images from different perspectives to obtain corresponding points,and finally obtains spatial point three-dimensional information through geometric principles.The paper mainly studies several key technologies of binocular stereo vision.The main contents include camera calibration,image feature extraction and matching,and three-dimensional reconstruction.In the calibration part of the camera,the imaging model of the camera is studied,and the relationship between the calibration coordinate systems is analyzed in depth.The internal and external parameters of the binocular camera are obtained by Zhang Zhengyou's calibration algorithm.In image feature extraction and matching,a SURF(accelerated robust feature)and LDB matching algorithm is first proposed.The algorithm first uses the Hessian matrix to extract the image feature points,rotates the coordinate axes to the main direction of the feature points determined by the Harr wavelet response in the circular neighborhood to avoid the computational cost of image rotation,and then uses the LDB calculation proposed in this paper.Sub-feature to describe the texture information of neighorbood of the feature point,forming a 252-bit feature vector.The experimental results show that the description method can effectively reduce the computational complexity of the descriptors and make the algorithm significantly improve in matching accuracy and matching time.Secondly,a new SIFT(scale invariant feature transformation)image matching algorithm that combines color and illumination information is proposed.The new algorithm first obtains the color compensation amount and the illumination compensation amount of each pixel of the color image,and increases the color compensation amount and the illumination compensation amount to enhance the contrast and reduce the matching error caused by the color difference when the image is grayed out;Grayscale image matching using SIFT algorithm.The experimental results show that the new algorithm can effectively distinguish the regions with different colors but similar gray levels,increase the matching points and accuracy of SIFT algorithm,and improve the matching performance of the algorithm.3D reconstruction section.After matching the feature points of the stereoscopic image pair,the corresponding relationship between the matching point pairs and the three-dimensional object points can be established by the calibrated internal and external camera parameters,and the three-dimensional information of the spatial points can be obtained.Finally,use Open GL's 3D display technology to display the reconstruction results.