【作者】 吴冠勇；

【导师】 韦穗；

【作者基本信息】 安徽大学 ， 电路与系统， 2004， 硕士

【摘要】 本文从计算机视觉中的多视图几何理论出发，主要讨论摄像机内参数估计技术，包括基本矩阵、摄像机自标定等问题的算法与实验。主要工作如下： (1)基本矩阵的鲁棒算法，它是摄像机标定的基础。首先介绍了几种常见的求基本矩阵的线性算法；然后给出基本矩阵的6点随机抽样一致算法—Random Sample Consensus(RANSAC)。 (2)基于SVD分解的摄像机自标定。给出了基于基本矩阵SVD分解(Singular Value Decomposition—奇异值分解)的摄像机自标定技术，首先介绍了通过对基本矩阵进行SVD分解所推出的Kruppa方程的简化形式，并将摄像机内参数的估计变成一个非线性优化问题，简化过程中无需用到绝对二次曲线，然后利用共轭梯度法来迭代得到的代价函数，最终求得摄像机内参数。 (3)运动参数约束下的摄像机自标定，也即人们所说的基于主动视觉的标定。首先介绍了一种基于极点信息的摄像机自标定方法，然后给出一种新的运动参数约束下的摄像机自定标方法，该方法的主要特点是可以唯一线性求解摄像机的5个内参数。最后给出实验结果，并验证了摄像机在纯平移运动下求解基本矩阵的2点算法的稳定性。更多还原

【Abstract】 In this paper, we mainly discuss the technology of estimating the camera intrinsic parameters based on multiple view geometry, which includes the algorithms and experiments of fundamental matrix, camera self-calibration, etc. The main work is as following:(1) Robust algorithms for fundamental matrix. It is the foundation of camera calibration. We firstly introduce several common linear algorithms of fundamental matrix, and then we give the algorithm named as six point Random Sample Consensus (RANSAC) algorithm.(2) Self-calibration based on the Singular Value Decomposition (SVD) of fundamental matrix. The proposed method relies on the Singular Value Decomposition of fundamental matrix, which leads to a particularly simple form of the Kruppa equations optimized by conjugate gradient method. The derivation doesn’t need the somewhat non-intuitive geometric concept of the absolute conic. At last, we give the result of experiments.(3) Self-calibration under the constraints of motion parameters. It is usually equal to the self-calibration based on active vision. We first introduced a self-calibration algorithm based on the epipoles, Then, we give a new algorithm under the constraints of camera motion parameters, it can get the 5 intrinsic parameters of the camera linearly and uniquely, at last, we give the result of experiment, and verify the stability of 2-point algorithm for fundamental matrix.更多还原