基于三视图几何约束的摄像机相对位姿估计

针对从影像中恢复摄像机位姿的问题，提出基于三视图几何约束的位姿估计算法.利用强制选择机制对影像进行特征点提取，通过光流法和前后向误差实现相邻三帧影像的特征点匹配；推导基于本质矩阵优化分解的位姿估计，利用对极几何约束构建目标函数，通过迭代优化确定旋转矩阵的唯一解，优化了唯一解的确定方法，提高了相对位姿估计效率，得到第1、2摄像机矩阵；基于三视图的几何约束关系，由三焦点张量和匹配特征点建立目标函数，由迭代过程得到第3摄像机相对于第1摄像机的位姿参数.结果表明，提出算法的鲁棒性、精度以及算法效率均优于传统算法，能够快速、准确地估计摄像机相对位姿，可以实现对旋翼无人机的轨迹跟踪.

Camera pose estimation based on three-view geometry constraint

A novel pose estimation method based on three-view geometry constraint was proposed in order to solve camera pose estimation from image sequence. Image feature points were detected by using brute force selection mechanism, and feature points were matched based on optical flow algorithm and forward-backward error. Then pose estimation based on improved essential matrix decomposition was deduced. Object function was established according to epipolar geometry constraint, and rotation matrix was iteratively computed. Then a unique solution was obtained. The first and second camera projective matrices were obtained. The object function was built based on trifocal tensor and matching points, and the third camera pose matrix that was relative to the first camera was iteratively computed. Results show that the proposed method improves the robust, position accuracy and efficiency. The method can achieve a precision estimation fast and estimate the trajectory of rotor UAV.