Kalman-based continuous pose estimation network for spacecraft

With the increasing number of space missions, the quantity of spacecraft and space debris has surged dramatically. The technology for on-orbit servicing (OOS), applied in space debris removal, retrieval of defunct spacecraft, rendezvous and docking, has developed greatly in the field of aerospace. The real-time six degree-of-freedom pose estimation of space objects holds significant value in on-orbit tasks such as spacecraft rendezvous and docking, on-orbit capture and servicing, as well as space debris detection and removal. In this paper, a continuous space-craft pose estimation method based on Kalman filtering and Long Short-Term Memory (LSTM) networks is proposed. A one-stage pose estimation network is designed for a single-frame image, employing a multi-scale structure to enhance pose estimation accuracy. Additionally, an LSTM network considering Kalman filtering is introduced, which explores temporal information to further improve the continuity and robustness of pose estimation. The proposed method is validated on both simulated and experimental datasets. Experimental results demonstrate certain advantages of continuous pose estimation method in terms of pose estimation accuracy and continuity.