Vision-based aircraft pose estimation with dual attention module for global feature extraction in complex airport scenes

ddressing the intricate task of aircraft pose estimation in challenging airport environments, we introduce a dual attention module (DAM) inspired by the human visual system. Comprising two innovative sub-modules, the Inverse Parallel Multi-Spectral Channel Attention Module (IPMS-CAM) and the Gaussian Spatial Attention Module (G-SAM), the DAM collaboratively captures global aircraft features. Specifically, the IPMS-CAM employs the discrete cosine transform and integrates attention mechanism to adapt to different levels of visual cues, effectively addressing the “where to look” challenge within complex airport scenes. In contrast, the G-SAM emulates the spatial attention process of the cerebral cortex, effectively tackling the “what to see” problem, and thereby emphasizing pertinent information while intelligently filtering out extraneous details. Our proposed approach is rigorously evaluated through experiments on two distinct airport surface datasets. The results substantiate the compelling competitiveness of our method when juxtaposed against prevailing attention mechanism techniques. Our approach achieves a noteworthy performance advancement of approximately 3% over the baseline method, all while incurring negligible computational overhead.