Pose-Induced Error Compensation in Monocular Vision for Long-Range Air Draft Estimation

Accurate air draft estimation is essential for safe maritime transportation. Conventional solutions mainly rely on laser sensors or binocular vision systems, which suffer from point scattering at long distances and require wide baselines, limiting their sensing performance. To address these limitations, this paper proposes a telephoto monocular-based framework for long-range air draft estimation. We realize this system by overcoming two technical challenges, including image degradation in long-range telephoto views and image distortion from the camera viewpoint, through two dedicated designs. The first design introduces a ship-specific attention mechanism that embeds domain-specific features of ship geometry to enhance long-range ship contours in degraded telephoto images. The second design develops learning-based pixel-wise calibration and geometry-based pose estimation to correct image distortion and convert image height into metric air draft. We collect a real-world dataset comprising 5000 telephoto images over three months under diverse weather conditions at Ningbo Zhoushan Port. Experimental results show that the proposed method reduces air draft estimation error by 56.4% compared to state-of-the-art approaches.