R-OFFlow: An Optical Flow Motion Estimation Method Guided by Offset Fields for Self-Compensation of Ladder Jitter in Offshore Wind Turbine Boarding

The violent swaying of gangway when maintenance vessel dock at wind turbine tower poses a severe threaten to the safety of maintenance personnel. Therefore, ensuring the stable docking of gangway is of vital importance for safeguarding the maintenance personnel. However, marine-surface vehicles often experience large-scale irregular movements due to the coupling effects of wind, waves, surge, and other disturbances, posing unique challenges. Real-world factors such as parallax and depth variations, the breakdown of the brightness constancy assumption, and texture absence constrained the accuracy of current optical flow estimation methods. To solve this issue, an offset-field (OF)-based unsupervised optical flow learning framework called R-OFFlow algorithm is proposed. Specifically, the motion field module called OF is formed by integrating the gyroscope data and heave information at first. Second, the self-guided offset fusion (SGOF) module is designed, which fuses the background motion extracted from the OF with the optical flow. Third, the feature extraction and optical flow refinement methods by recurrent all-pairs field transforms (RAFTs) are integrated to enhance the feature extraction. The combination of the above modules guides the network to focus on motion details, thus achieving high-precision unsupervised optical flow learning. Furthermore, a dedicated dataset called wind power optical flow (WPOF) is constructed. Finally, the comparative analysis experimental results demonstrated that the proposed method improves by at least 12.12% compared to other algorithms in various complex offshore scenarios, outperforming state-of-the-art methods. Ablation studies further validate the effectiveness of the modules in the proposed algorithm.