Vision-based Navigation Systems for Autonomous Spacecraft Rendezvous: A Comprehensive Review

Abstract

Autonomous spacecraft rendezvous represents one of the most challenging navigation problems, requiring precise relative state estimation in the absence of GPS, under extreme lighting conditions, and with strict computational and power constraints. This comprehensive review examines vision-based navigation systems that enable autonomous approach and docking operations for on-orbit servicing, debris removal, and formation flying missions. We systematically analyze the evolution of vision-based rendezvous techniques from early marker-based systems to modern learning-based approaches, categorizing methods by their sensing modality: monocular cameras, stereo vision, time-of-flight cameras, and multi-spectral imaging. The review examines pose estimation algorithms spanning feature-based methods (SIFT, SURF, ORB), model-based approaches utilizing known target geometry, and deep learning techniques including convolutional neural networks for keypoint detection and pose regression. Particular attention is devoted to handling the unique challenges of space environments: extreme illumination variations during orbital day-night transitions, lens flare from direct sunlight, motion blur during rapid relative motion, and the lack of atmospheric scattering that produces harsh shadows. We evaluate robustness to non-cooperative targets including tumbling satellites and debris without fiducial markers, analyzing shape-from-silhouette, photometric stereo, and structure-from-motion techniques. The review comprehensively examines sensor fusion architectures combining vision with inertial measurements, star trackers, and LiDAR for enhanced accuracy and reliability, assessing Kalman filtering variants, particle filters, and factor graph optimization approaches. Computational efficiency is critically analyzed, comparing algorithm runtime and memory footprint against available spacecraft processors, with emphasis on hardware acceleration through FPGAs and specialized vision processors. We synthesize validation methodologies including hardware-in-the-loop testbeds with robotic manipulators, proximity operations simulators, and on-orbit demonstration missions, identifying gaps between laboratory performance and space-qualified systems. The review examines emerging technologies including event-based cameras for high dynamic range imaging, neuromorphic vision processors for ultra-low power operation, and AI-based anomaly detection for fault tolerance. Application-specific considerations are discussed for different mission profiles: large cooperative spacecraft with docking ports, small satellites with limited computational resources, and debris with unknown or damaged surfaces.