空间翻滚非合作目标相对位姿估计的视觉SLAM方法

针对空间翻滚非合作目标相对位姿测量中目标先验信息缺失和模型不确定问题，将移动机器人视觉同步定位与建图（SLAM）贝叶斯滤波估计模型推广到非合作目标相对位姿测量中，提出一种基于视觉SLAM的翻滚非合作目标相对位姿估计方法。首先，构建了相对位姿估计的贝叶斯滤波状态转移模型和量测更新模型。其次，为避免平动噪声协方差过大导致滤波性能下降的问题，对状态转移方程进行优化，采用最小二乘估计方法预测位置参数。最后，采用一种联合无损卡尔曼滤波和粒子滤波的贝叶斯滤波方法实现了目标全部位姿参数的快速平滑估计。基于计算机合成图像数据和真实图像序列的仿真实验表明：提出的方法具有较优的速度和精度，且对目标速度变化、特征提取和数据关联误差等具有较好的鲁棒性。

Relative Pose Estimation of Space Tumbling Non cooperative Target Based on Vision only SLAM

Considering the space tumbling non-cooperative target motion uncertainty and without any priori knowledge, a relative pose estimation method based on vision-only simultaneous localization and mapping(SLAM) for space tumbling non-cooperative target is proposed. Firstly, the state transition equation and measurement update equation of Bayes filter model for relative pose estimation is established. Secondly, in order to overcome the detrimental effects caused by large noise covariance of translation process model, the least square estimate is adopted to predict the position parameter and a improved process model is obtained. Finally, a fast and smooth relative pose estimation is realized by using a Bayes filter combining unscented Kalman filter(UKF) and particle filter(PF). Results from both synthetic and real image sequences show that the method is accurate and efficient, and is robust to the changeful target rotation velocity and the errors of feature extraction and data association.