Visual Positioning of a Satellite for Efficient On-Orbit Docking

This paper tackles the challenge of visual positioning for fuel pipelines in satellite refueling missions within a space environment, proposing a novel method based on 3D Gaussian splatting scene representation. The method formulates pose estimation as an optimization problem, generating an objective function through 3D Gaussian splatting rendering, and solving it iteratively using optimization techniques such as gradient descent, thereby achieving high-precision pose estimation. To improve the stability of the optimization process, the paper introduces color transfer technology to mitigate color discrepancies and employs an early-stopping strategy to balance positioning accuracy with computational efficiency. The method is validated through simulation experiment, real-world prototype experiment, and real-world application experiment. The results demonstrate that, compared to traditional visual positioning techniques, this approach offers superior accuracy in fuel injection hole positioning tasks.