EvSat3D: Satellite Pose Estimation and 3D Reconstruction With Event Camera

摘要

Accurate object pose estimation is vital in robotics and satellite operations, especially during rendezvous missions involving proximity interactions with orbiting satellites, non-cooperative objects, or space debris. Traditional frame-based cameras struggle in the space environment due to limitations such as motion blur, low dynamic range, and low frame rates—conditions exacerbated by rapid relative motion, abrupt lighting changes, and high contrast variations in orbit. Event cameras offer a compelling alternative with their unique advantages: ultra-high temporal resolution (>10,000 fps), wide dynamic range (up to 140 dB), and low power consumption (1mW), making them especially suited for the extreme visual conditions in space. This paper introduces EvSat3D, a comprehensive and robust pipeline for event-based satellite pose estimation and 3D reconstruction. The pipeline leverages event-to-frame (E2F) conversion algorithms to generate frames from asynchronous event data and applies Structure-from-Motion (SfM) techniques for accurate camera pose recovery and 3D scene reconstruction. We validate EvSat3D using the only publicly available event-based satellite dataset and demonstrate its effectiveness under challenging conditions. Our results highlight the feasibility and performance benefits of using event cameras for satellite navigation and pose estimation tasks in space.