基于人工标志的单目视觉下无人机位姿估计

摘要

就无人机的自定位问题,文中提出一种数字图像处理结合2D-3D场景重建的位姿估计算法.在已知人工标志世界坐标的前提下,对于单目视觉下机载非线性相机拍摄的单帧图像,利用图像处理技术为位姿解算提供特征点数据.位姿解算提供了一种改进的N点透视算法,该算法结合正交迭代算法,以N点透视算法计算的旋转矩阵值作为迭代算法初值,迭代得到最终估计结果.通过仿真实验和实物实验验证了算法的可行性.文中提出的算法,较N点透视算法,提高了位姿估计的精度且收敛速度快,给无人机室内位姿估计提供了的一种新的解决思路.%This article proposes a pose estimation algorithm with a combination of digital image processing and 2D-3D scene reconstruction to deal with UAV self localization. Under the premise of the known world coordinates of the artificial mark , we obtain the information from the images taken by a nonlinear onboard camera and take some methods of digital image processing to provide feature point datas for subsequent calculation. Pose estimation provides a improved N-point perspective algorithm which combines orthogonal iteration. N-point perspective calculates attitude angles matrix as the initial values of orthogonal iteration. The feasibility of the algorithm is verified by simulation experiments and real experiments. Compared with N-point perspective algorithm, this method improves the pose estimation accuracy and fast convergence and provide a new idea for the UAV indoor pose estimation .