Visual robustness under image degradations for planetary exploration

Abstract

Robotic space exploration demands reliable visual perception systems capable of operating in environments where acquisition conditions can severely degrade the available visual information. This work presents a systematic evaluation of visual feature extractors — both classical and learning-based — under three degradations representative of planetary exploration scenarios with constrained sensing: spatial resolution reduction, radiometric quantization, and photoelectronic degradation associated with low illumination. The evaluation is conducted on navigation sequences in planetary environments using metrics encompassing geometric consistency, descriptive robustness, and computational cost. Results show that no universally robust extractor exists, as performance is closely tied to the type of degradation: spatial reduction limits the number of useful correspondences, radiometric quantization partially preserves geometry while reducing discriminability, and low illumination undermines correspondence stability. This study provides actionable criteria for selecting visual feature extractors in robotic planetary exploration missions operating under sensing constraints.