Research On 3D Reconstruction Technology Based On MonoCular Vision

At present, with the rapid development of machine vision and 3D technology, and the great progress of human science and technology, three-dimensional reconstruction technology is increasingly becoming one of the hot focus of research in computer vision. three-dimensional reconstruction Based on image can be attributed to: get image matching points set from the image feature points extraction and matching, and get the outside parameters of the camera by calculating the essential matrix and singular value decomposition of it, and obtain the camera internal parameters from camera calibration. Then obtain the camera projection matrix by the external and internal parameters. According to the relationship between the perspective projection and the least square method, we can recover the 3D coordinates of the object in the space coordinate system from the camera projection matrix and the matching point pair, and then reconstruct a three-dimensional model of the object. With respect to the binocular stereo vision, monocular stereo vision has characteristic of simple structure, flexible movement, easy calibration, low cost and so on. Especially now that the popular of phones and tablets, relying on a single camera reconstruct a particular target become a new hot spot, and is becoming a new demand.This paper is use a cell phone camera to shoot the object from different angles to obtain image sequence of the object from different angles, and then reconstruct a three-dimensional model of the object through the camera calibration, feature point extraction and matching, 3D reconstruction and other steps. Three-dimensional reconstruction technology has important applications in industrial automation, medical assistance, virtual reality, visual navigation, SLAM, and other fields.Firstly, introduce the camera imaging model, coordinate transformation, epipolar geometry, fundamental matrix, essential matrix, camera calibration and other basic theory, and analyze detailedly the mathematical derivation of the basic matrix, the essential matrix and the extrinsic parameters of the camera. And analyze detailedly of the current mainstream of the camera calibration method, found Zhang Zhengyou plane calibration method avoid the shortcomings of traditional calibration of equipment requirements, complicated operation and low calibration accuracy, and have advantages of high calibration accuracy, robustness and suitable for a lot of applications, etc compared to self-calibration method. Therefore, this paper use Zhang Zhengyou plane calibration method based on OpenCV, and programming the camera calibration process in the analysis phase of the experiment, which achieved satisfactory calibration results.Then, study intensively Harris algorithm, SURF algorithm, after RANSAC algorithm in the image feature extraction and matching algorithm, combined with the algorithm described above, proposed and elaborated Harris-SURF feature extraction and matching algorithms. Harris-SURF algorithm is the improvement of Harris algorithm and SURF Algorithm. Harris-SURF feature extraction and matching algorithm not only have characteristic of simple calculation, extraction of the feature points is the corner of the Harris corner detection algorithm, also have scale invariant features of surf algorithm. The experimental results show that the Harris-SURF algorithm improves the real-time performance and the high efficiency of the SURF algorithm.Next, combined with the basic theory first of few chapters, proposed a three-dimensional reconstruction program based on monocular vision. First elaborated the basic principles of three-dimensional reconstruction, gived the flow chart of three-dimensional reconstruction program based on monocular vision, and brief introduced the process of each step, then describe the method of solving the camera projection matrix. And elaborated the three-dimensional coordinate space point algorithms by projection matrix and matching point set of last chapter. Experimental results show that the three-dimensional spatial coordinates of the point to be computed by the reconstruction of the object is correct.Finally, programming developed a three-dimensional reconstruction of an application based on monocular vision by using OpenCV and OpenGL programming language in Qt development environment on Windows. which implemented successful the camera calibration, image feature extraction and matching, calculation of three-dimensional spatial coordinate points of the object to be reconstruction, reconstruction of three-dimensional point cloud model space sparse of the object to be reconstruction. The results have proved the feasibility of 3D reconstruction scheme based on monocular vision proposed in this paper.