空间失效慢旋卫星视觉特征跟踪与位姿测量

空间碎片或者失效卫星往往绕其惯性主轴做自旋运动，虽然处于一种稳定状态，但是对其运动状态测量的难度相比三轴稳定目标而言又增加了许多，主要体现在复杂光场环境下和目标观测面周期性进出视场引起的特征点丢失、尺度变化和旋转变化，以及长时间连续观测引起的累积误差增大、位姿解算不收敛等难题。通过构建优化目标特征数据库，将传统测量方式中前后帧的匹配转变为当前帧与特征数据库的匹配和特征点的优化，即使中间过程帧出现了偏差仍然可以较好地跟踪后续图像帧的正确位置。在李代数空间下建立位姿向量微分扰动方程，获得测量值与估计值的残差目标函数，基于贝叶斯法则最大后验概率和李群与李代数的对指变换法则，求取位姿向量最优解。解决了李群空间下由于位姿变换矩阵不可加性而无法优化的问题，提高了连续测量过程中系统的测量精度。实验结果表明，无优化测量方法无法保证整个旋转周期的有效测量；通过增加位姿优化过程，连续稳定测量时间明显延长，针对以12（°）/s自旋运动的目标，测量稳定段误差在2°以内，旋转角速度测量误差为0.12（°）/s。

Visual feature tracking and pose measurement for slow rotating failure satellites

Space debris or failed satellites often spin around their main inertial axis. Despite their stable state, measurement of their motion state is more difficult than that of a three-axis stable target. Difficulties mainly involve loss of feature points, scale changes and rotation changes in a complex light field environment, and those caused by periodic entry and exit of the target observation surface, as well as increase of cumulative errors and non-convergence of the calculated position and attitude resulted from long-term continuous observation. An optimized target feature database is constructed to turn the matching of the front and back frames in the traditional measurement method into the matching and optimization of the feature points of the current frame and the feature database. In this way, even if an error occurs in the middle process frame, the correct position of the subsequent image frame can still be tracked fairly well. The differential perturbation equation of the pose vector in the Lie algebra space is established to obtain the residual objective function of the measured and estimated values. Based on the maximum posterior probability of the Bayes' rule and the pair-finger transformation rule of Lie group and Lie algebra, the optimal solution of the pose vector is obtained, thereby solving the non-optimizable problem in the Lie group space due to the non-additivity of the pose transformation matrix, and improving the measurement accuracy of the system in the continuous measurement process. Experimental results show that the unoptimized measurement method cannot guarantee effective measurement of the entire rotation cycle, and that increasing the pose optimization process significantly extends the time of continuous stable measurement. For targets with spin motion of 12 (°)/s, the measurement error of the stable segment is within 2°, and that of the rotation angular velocity is 0.12 (°)/s.