| dc.description.abstract | Missions to explore planetary bodies (e.g., Moon, Mars, Titan, Europa, Enceladus) through surface exploration have been planned for the next decades. These missions depend on autonomous optical navigation capabilities for safe entry, descent, and landing near key scientific areas of interest, potentially near hazardous terrain. Terrain Relative Navigation (TRN) enables autonomous precision landing by matching descent images to an a priori orbital map. However, performance degrades significantly when large differences in solar illumination exist between the map and descent imagery, particularly under high azimuth angle changes, due to terrain-induced shading inversions that break assumptions of photometric consistency in both frequency and intensity-based correlation methods. This thesis presents a set of methods to robustify TRN performance under large directional illumination changes, addressing three core challenges: understanding the failure of existing TRN methods, improving feature matching robustness to varying Sun angles, and generating reliable navigation maps from incomplete orbital imagery. First, the physical cause of correlation failure is characterized through a frequency-domain analysis of shading effects. Building on the azimuth impact matrix, which models the directional dependence of shading-induced phase reversals, this work applies it as a frequencydomain correction to frequency correlation to improve correlation peak accuracy across large solar azimuth differences. Second, a novel frequency-domain photometric correction method, Solar Orientation Layering via Frequency Image eXtraction (SOLFIX), is introduced to produce corrected map products aligned with expected descent conditions by layering spatial frequency content aligned with known solar angles from multi-resolution, multi-illumination orbital imagery. Third, a predictive illumination-aware map is developed to identify terrain regions likely to yield reliable correlations. This map integrates solar azimuth geometry, terrain aspect from low-resolution digital elevation models, and spatial frequency information to pre-filter unreliable areas of the map prior to localization. The proposed methods are validated on Mars orbital datasets, simulated terrain renderings, and NASA JPL’s field test imagery, each with Sun angle variations. Despite wide Sun angle differences, the proposed methods recover accurate localization where baseline TRN methods fail due to the bias from shading inversions caused by large solar azimuth differences. These contributions enable reliable optical navigation in scenarios where acquiring new orbital maps is infeasible and support future planetary missions operating under severe illumination constraints. | |
