为了更加实时、精确地实现小型无人机视觉导航，利用矩形的几何特性以及空间点、线的共面特征，提出了一种基于单目视觉的位姿分步估计方法。该方法在任意位置对空间矩形进行拍摄，通过获取的一对正交消隐点，确立无穷远点在像平面的投影关系，完成姿态估计；利用光心、空间矩形以及其投影的共面性特征建立约束方程，求解空间矩形在摄像机坐标系下的法向量，结合欧氏空间线性变换的不变性，实现摄像机坐标系中矩形4顶点坐标的线性求解，并根据空间点在世界坐标系与摄像机坐标系间的转换关系，完成位置估计。为了抑制图像噪声对位姿估计的影响，建立基于空间点、线共面特征的指标函数，利用NM寻优算法实现对位姿参数的非线性优化。实验结果表明，该算法具有计算精度高、实时性强、适用范围广的优点，设计的优化算法能够有效地抑制图像噪声，提高了位姿估计的稳健性。

In order to improve the vision navigation of unmanned aerial vehicle more real-timely and accurately, a pose estimation method step by step based on monocular vision is proposed using rectangular geometry feature and coplanar feature of points and lines. This proposed method shoots the rectangular target under any position, and two orthogonal vanishing points are achieved. The constraint equation of projection between infinite points and orthogonal vanishing points are established to calculate the rotation parameters. And also a constraint equation is established by using the coplanar feature of camera optical, rectangular target and its projection, which are used to solve the normal vector of rectangular target under camera coordinate system. Based on the invariance feature of linear transformation under Euclidean space, the four vertices of rectangular target under camera coordinate system can be linearly calculated and the solution of the translation estimation can be accomplished through the translation between camera and world coordinate system. Aiming at restraining the influence of noise to estimation, the indicated function based on the coplanar feature of points and lines is obtained, using the NM optimization algorithm to realize the nonlinear optimization of pose parameters. Experimental results show that the proposed method has the advantages of high precision, real-time and wide application. The robustness of estimation algorithm is improved by reducing the image noise through parameter optimization.