Model-free 6D pose estimation for noncooperative space targets driven by implicit surface anchors

Abstract

Precise 6D pose estimation of non-cooperative space targets is a core enabling technology for on-orbit servicing. Existing methods largely rely on predefined CAD models, which often suffer significant performance degradation when deployed in space environments characterized by extreme depth variations. In this paper, an end-to-end 6D pose estimation framework based on implicit surface anchors is proposed. By fusing cross-modal features extracted from RGB and depth images, the framework autonomously identifies rigidly invariant 3D anchors within the target's local canonical space. Precise pose resolution is achieved through a differentiable BPnP layer. Furthermore, a multi-constraint joint optimization system is employed to prevent anchor pattern collapse, and is combined with a cosine annealing strategy to stabilize training dynamics. This approach enables accurate 6D pose estimation without the need for predefined CAD models. Simulation results on public datasets demonstrate the effectiveness and robustness of SoPD-net.