Feature Point Extraction and Descriptor Matching Method Based on Phase Coherence and Direction Map

During orbital operations, targets experience significant illumination variations due to their periodic shadow transitions resulting from their trajectory around the Earth. Coupled with the diverse radiometric properties of target surface materials, particular illumination angles can induce phenomena such as specular reflections from certain coatings or significant radiant flux absorption by others. These combined effects frequently lead to either over-saturated or critically under-exposed regions within captured images, thereby posing substantial impediments to the robustness of vision-based systems employed for on-orbit tasks. Aiming at the problem of degradation of the performance of traditional feature point detection algorithms due to extreme illumination variations of spatial targets in on-orbit vision tasks, this study proposes an illumination robust feature detection method based on phase consistency. The structural feature extraction is achieved through the construction of a multi-scale phase consistency saliency map. This process integrates directional consistency verification with soft non-extreme value suppression to enhance feature stability. Additionally, a two-phase feature descriptor with rotational invariance has been designed. The experimental results show that: under the simulated overexposure (f=2.0) and overdarkness (f=0.1) conditions, the method in this paper detects an average of 2609 feature points, which is 195.5% and 74.6% higher than that of Speeded Up Robust Feature(SURF) and Features From Accelerated Segment Test(FAST), respectively; the number of effective matching pairs reaches 714 pairs, and the relative degradation is only 11.3% under the extreme illumination conditions, which is significantly better than the traditional grayscale-dependent method. The phase consistency method effectively overcomes the feature degradation problem caused by the linear scaling of illumination through the frequency domain phase information, which provides a reliable technical support for the visual navigation of spatial targets.