A monocular vision-based relative position and attitude measurement algorithm was presented,which combined the reconstruction of targets with the measurement of relative positions and attitude for non-cooperation targets. Firstly,a series of points on the targets were reconstructed using the images obtained in the earlier measurement with a fast and globally convergent structure and motion estimation. Secondly, the relative position and attitude were calculated by using a Kalman filter based on recursive depth parametrization,in which new points observed from new images were recovered and added to the 3D point set to improve the accuracy and robustness of the measurement. Experiment result shows that the attitude measurement can achieve an accuracy higher than 1°,and the position measurement can achieve an accuracy higherthan 2 cm,which indicates the high precision of the proposed algorithm.%为了解决非合作目标的相对测量问题,提出了一种基于单目图像序列目标重构结果的非合作目标相对位姿测量方法.该方法将目标的三维重构与相机的位姿信息计算相结合,首先利用观测前期得到的图像序列,通过非线性优化算法计算得到目标上部分三维点坐标;然后基于该三维点集合,建立递推深度模型,对相机的相对位姿信息和新观测到的目标点同时进行卡尔曼滤波估计.航拍测量试验表明,随着图片数量的增多,精确重构点的比例(重投影误差小于1个像素的点)不断提高,80%的图像中精确重构点比例优于89%;基于公共数据集的试验表明,该算法对姿态估算精度可达1°以内,位置测量的精度可达到2 cm以内.以上试验结果表明,该算法具有较高的测量精度.
展开▼
