A Novel Absolute/Relative Fusion Approach for Visual Localization of Planetary Lander Employing a Unified Error Metric

Absolute pose estimation via descent imagery and map matching is critical for precise planetary lander localization. Existing absolute localization methods suffer from trajectory uncertainty and unsmoothed reconstructions due to the image-to-map matching errors. This research proposes an absolute/relative fusion approach for visual localization of planetary lander. A unified error metric is designed by jointly optimizing reprojection, landmark consistency, and inter-frame pose errors, therefore converting all error metrics to a unified pixel scale to avoid weighting issues. This proposed approach ensures both precise localization and smooth trajectory generation. Validation includes lunar simulation data (with ground truth) and lunar analog UAV data, evaluated using localization deviation metrics. Moreover, the Chang'E-5 mission data is also utilized to evaluate the performance of the proposed solution. By Comparing with other typical localization methods, it demonstrates that the proposed approach significantly improves the localization accuracy and trajectory smoothness. This study offers a new solution for high-precision visual localization optimization with applications in descent trajectory recovery and landing point localization of planetary lander.