Research Of The3D Reconstruction Theory And Algorithm Based On Stereo Vision

In this paper we research3D reconstruction of key technologies based on stereo vision combining with practical projects and work.Formed a basic suit of3D reconstruction software,which can be used for practical iterm project.Stereo vision is an important branch of machine vision and computer vision, it covered the knowledge of fields such as artificial inteligence、computer science、image processing and pattern recognition.It has important practicality value,and is also very significant for promoting the research of human vision mechanism.In this paper,3D reconstruction of the existing camera calibration, stereo matching and3D reconstruction techniques have been studied and analyzed, in addition, studied the3D reconstruction model in open source computer vision library OpenCV and proposed combination of corner-based feature extraction algorithm sub-pix camera calibration method and Zhang Zhengyou new camera calibration algorithm. SIFT uses an improved algorithm to solve the problems,e.g scarcity of lunar surface texture, great impact to the the brightness of the image by the transformation angle of sun, and stereo matching difficult. Finally analyzed the3D reconstruction algorithm base on image feature points of calibration implemented all of the algorithms in this paper and analyzed the results of each data.Firstly, introduced three coordinate systems often used in the calibration process: image coordinate system, camera coordinate system and the world coordinate system. Then analyzed camera model and calibration thought. Focus on analyzing plane calibration method based on Zhang Zheng you camera calibration method, carried out experiment, proposed combination of corner-based feature extraction algorithm sub-pixel camera calibration method and Zhang Zheng you new camera calibration algorithm and got the internal and external camera parameters by the calibration results including the camera matrix, distortion factor, the rotation vector, rotation matrix and translation vector and so on.Secondly, we carry on feature extraction and stereo matching based on the feature points, which are the foundation and key of3D reconstruction. To analyze and contrast common basic feature point extraction operator Harris and SUSAN corner detection operator, and we take the experiment as verification. Taking use of the improved SIFT algorithm to resolve the sparsity of image texture on the lunar surface. For the problem about the image brightness is affected by sun transformation angle strongly and difficulty of stereo matching, we used lunar surface multi-spectral image data, different scales and different points of view of the lunar surface images to carry on feature extraction and stereo matching experiments separately.Finally, we analyzed the significance of image correction in the3D reconstruction and explained principle of image correction method and the algorithm. Focus on studying fast3D reconstruction algorithm of images which have been corrected. Contemporarily, we take the experiment to verify and analyze the3D reconstruction algorithm of images which have been corrected or uncorrected.In this paper, for all the algorithms, we took Visual C++, Open CV and Open GL as the platform to verify and analyze their results...