光场相机作为一种新型的成像系统，可以直接从一次曝光的图像中得到三维信息。为了能够更充分有效地利用光场数据包含的角度和位置信息，完成更加精准的场景深度计算，从而提升光场相机的三维重建的精度，需要实现精确的几何建模，并精确标定其模型参数。该方法从薄透镜模型和小孔成像模型出发，将主透镜建模为薄透镜模型，将微透镜建模为小孔成像模型，结合光场相机双平面模型，将每个提取到的特征点与其在三维空间中的射线建立联系，详细解释了内参矩阵中每个参数的物理意义，以及标定过程中初值确定的过程，并在镜头径向畸变模型的基础上进一步应用了相机镜头的切向畸变模型以及基于射线重投影误差的非线性优化方法，改进了光场相机的标定方法。实验显示，该方法的RMS射线重投影误差为0.332 mm，与经典的Dansereau标定方法相比，进行非线性优化后得到的射线重投影误差精度提升了8%。该方法详细分析的场景点与特定像素索引的推导过程对光场相机的标定具有重要的研究意义，为光场相机光学模型的建立与初始化标定奠定了基础。

When the collimator is placed horizontally and installed obliquely, its optical parameters will be greatly different due to different stress states. In order to accurately evaluate the focal length of collimator, according to the mapping relationship between the point on the focal plane of the collimator and the angle of the total station, an accurate mathematical model of the relationship between the focal length and the angle of the total station under the condition of oblique installation is established, the principle projection error caused by the rotation of the vertical axis of the total station is corrected. Several groups of data are collected by total station and experimental verification is carried out. After correcting the distortion, the focal length calculated by each testing point when the line segment is parallel to the vertical wire are 1 980.03 mm, 1 983.45 mm, 1 982.79 mm, the average focal length, i.e. the true value, is 1 982.09 mm. When the distortion is corrected but the projection error is not corrected, the focal length calculated from each testing point when the line segment is parallel to the horizontal wire of the reticle is 996.42 mm, 995.23 mm, 995.22 mm, the relative error of the average focal length is 50.2%. The range of focal length calculated by each testing point when the line segment is located in different quadrants and parallel to the horizontal wire of the reticle is 4.74 mm after correcting the projection error and distortion, the average focal length of all testing points is 1982.69 mm, the difference between the average value and the true value is 0.6 mm. The maximum relative error between the extended uncertainty of the focal length calculated by different testing point and the true value of the focal length is 0.36%. This value is far less than the stipulation in GB/T 9917.1-2002 that the relative error between the measured focal length and the nominal focal length in the photographic lens does not exceed ±5%. The experimental results show that the model has universality and high accuracy, the phase of the target slit in the reticle is allowed to be a random value, there is no need to adjust the slit to be strictly parallel to the vertical wire of the total station, the model has great engineering application value for the in-situ detection of the focal length of the collimator under the condition of oblique installation.

在自由双目立体视觉系统中，左右相机自由旋转，导致相机位姿发生实时变化。针对这一问题，提出一种重投影优化的自由双目相机位姿估计方法。通过分解相邻两幅图像间的单应矩阵估计相机的运动参数；将其作为初值计算重叠拍摄区域内特征点云的重投影误差，构建目标函数；利用非线性优化算法优化目标函数，得到最优的相机运动参数，结合相机旋转前的位姿参数，计算当前位置的相机位姿。仿真实验表明，位姿估计误差随重投影误差的减小而减小，所提算法能快速稳定收敛至全局最优值；水泥模型的三维重建实验表明，利用所提的相机位姿估计算法能够有效生成模型的三维点云，并可实现相邻点云的高精度拼接，拼接点云模型上两点距离的平均误差为1.68%。

针对有限或无重叠视场的多目相机系统，提出一种基于ChArUco(chess augmented reality university of cordoba)平板的多相机标定方法。采用ChArUco平板进行相机标定；基于双目重投影误差进行全局优化，为无重叠视场的相机间进行位姿传递提供精度保证；最后通过空间点重建进行性能测试。实验结果表明在点间距25 cm范围内平均测量误差为0.11%。

大多数视觉里程计通过跟踪图像序列中点特征几何位置的变化实现对相机位姿的估计。线是点的集合，相对于离散的点，帧间线特征的位置变化更具有显著性，因而有利于提高特征检测跟踪的鲁棒性。另外，在一些弱纹理场景中，点特征不够丰富，作为对点特征的补充，提出了一种融合点线特征的立体视觉里程计算法。构建新颖的点线重投影误差模型作为目标函数求解旋转矩阵和平移向量。模型中，使用Huber核函数减小特征误匹配对优化过程的影响。选取ORB算子检测点特征，LSD算子检测线特征，匹配时施加恒速约束、环形匹配、Bucketing约束和外观几何约束，提高特征匹配的速度及精度。采用公共数据集KITTI和EuRoC对算法进行评测，实验结果表明，该算法在多种场景中的鲁棒性能，相较于其他具有代表性的视觉里程计算法，在精度方面有提升。