Real-Time Vision-Based Pose Tracking of Spacecraft in Close Range Using Geometric Curve Fitting

Abstract

This article presents a new framework of real-time vision-based pose tracking for spacecraft in close range using geometric fitting of the imaged geometric primitives (GPs) on the spacecraft. At the first time instant, the tracking is initialized with the template-based pose retrieval and GP-based pose determination. At each subsequent time instant, with the pose prediction from the extended Kalman filter (EKF) as initial value, the GPs are associated with the corresponding image data, and thereby the maximum-likelihood estimation (MLE) for spacecraft pose can be obtained in real time by geometrically fitting the GP projections over the corresponding image data with generalized expectation–maximization and M-estimation. Using the MLE, the EKF generates the final pose estimation and predicts the pose at the next time instant. The basic configurations of the GPs are investigated for the stability of tracking. Sufficient experiments validate the accuracy and the real-time performance of the proposed method.