基于视觉的自动空中加油近距相对位姿估计

Vision based close-range relative pose estimation for autonomous aerial refueling

针对无人机自动空中加油过程中近距离高精度相对位姿估计问题,提出了一种基于特征跟踪的单目视觉方法。在加油机机身预先设置多个辅助识别标志点,安装在无人机机头上的摄像机获取加油机图像,将提取出的图像特征点与根据摄像机成像模型计算出的预设标志点坐标进行匹配,剔除匹配粗差后迭代估计出加油机和无人机的相对位姿信息。基于Vega Prime软件建立了三维视景仿真系统。仿真结果表明:该方法能够提供6自由度的相对位姿信息,位置分量能够满足自动空中加油相对位姿估计的精度要求,姿态角的平均绝对误差不超过1°,并且对特征点丢失或错误匹配具有较好的容错性。

A feature-tracking based monocular vision method was developed for close range, high-precision relative pose estimation, a key issue in autonomous aerial refueling （AAR） for unmanned aerial vehicles （UAV）. A number of beacons were placed on the body of the tanker as pre-set feature points, with a camera installed on the nose of the UAV to capture images of the tanker. Feature points were extracted and matched with coordinates derived from the camera imaging model. After elimination of gross matching errors, the relative pose information between the tanker and the UAV was calculated by an iterative algorithm. A three dimensional visual simulation system based on Vega Prime showed that this method could provide six degrees of freedom information on the relative position and orientation. The position measurements have the required accuracy, while the mean absolute measurement error of the attitude angle is less than 1 degree. This method meets the accuracy requirement of AAR and has good fault-tolerance for loss or mismatch of feature points.