Relative pose estimation with efficient feature distribution in landing guidance

Relying on vision forfeature keypoint extraction, matching, and Perspective-n-points (PnP)calculation to achieve relative pose estimation is an effective solution forcarrier-based aircraft to achieve high-precision autonomous landing. Existingmethods typically only perform global consistency feature detection withoutverifying the reliability of the detected features or considering the featuredistribution of key points. To address this issue, we propose a carrier-basedaircraft landing guidance state estimation algorithm based on efficientdistribution sampling of deep features. By extracting key points with highstate values, the relative pose estimation of the landing process is achieved.A multi-level feature coupling is constructed to establish a variable scale keypoint localization mechanism for key point extraction and detection. Anattention mechanism module is constructed to evaluate the state of highreliability key points. The experimental results show that the proposed methodhas better robustness and relative pose estimation accuracy, achieving robustand high-precision online relative pose measurement in airborne monocularvision landing guidance.