Multi-Sensor Fusion-Based Relative Motion Parameter Estimation for Non-Cooperative Targets

Accurately determining the relative pose and motion of non-cooperative targets during proximity operations, including on-orbit servicing and rendezvous and docking, remains both critical and technically demanding. This paper proposes an estimation framework that integrates stereo vision and LiDAR measurements: stereo vision provides continuous local feature constraints, while LiDAR measurements supply global pose constraints. These two modalities complement each other, enhancing the observability of the system and suppressing drift. At the filtering level, this study employs an Adaptive Error State Kalman Filter (AESKF) that recursively adjusts the measurement noise covariance based on innovation statistics, introduces a regularized projection via eigendecomposition to enforce positive definiteness, and incorporates a forgetting factor to tune the filter’s dynamic response and stability, and incorporating the dynamics of the non-cooperative target. Numerical simulation results demonstrate that the proposed fusion scheme achieves significantly higher accuracy in motion parameter estimation than any single sensor modality.