Relative state estimation of model-unknown spinning noncooperative target using stereo EKF-SLAM

A new method for real-time relative state estimation of noncooperative target using stereo vision is proposed. This mainly deals with spinning target without any prior information known and can be applied in many space missions, such as on-orbit servicing, space debris mitigation, etc. The method is inspired by the popular SLAM algorithm in computer vision community which is used for locate a robot in unknown scene. An EKF (extended Kalman filter) based method is chosen for relative state and model parameter estimation and map building due to the limited onboard computing resource. The algorithm consists of three parts: (1) geometric reference frame selection, which formulate the direct measurement for stereo vision; (2) localization and mapping, which reconstruct the features on target's surface and track pose between serial frames; (3) model parameter estimation, which estimate the inertial axes and center-of-mass (CM) of target. Results from numerical simulation demonstrate the performance and viability of the proposed algorithm.