Long-term detection and tracking algorithm for moving vessels by maritime UAVs
An algorithm for long-term maritime target detection and tracking based on the combination of YOLOv5 and ECO_HC is proposed to address the problem of tracking failure caused by occlusion of ship hulls and ships leaving the field of view during unmanned aerial vehicle(UAV)tracking of ship at sea.First,perceptual hashing and the ratio between the second and first major modes are used to comprehensively assess the reliability of the tracking process.In the event of target loss,the YOLOv5 detector is utilized to reposition the target and initialize the tracking model.Thereby,the accumulation of erroneous information is eliminated.Secondly,to address the rotational changes of the target during tracking,the Fourier-Mellin transform is employed for rotation parameter estimation,mitigating performance decline due to target rotation.The proposed algorithm achieves an average precision and success rate of 83.9％and 76.7％,respectively,on the OTB-100 dataset.Field experiments of ship tracking in actual maritime scenarios on UAV platforms show precision and success rates of 80.9％and 60.4％under complete occlusion,and 90.2％and 48.3％when the target is out of the field of view.The experiments demonstrate that the proposed algorithm can effectively suppress the influence of common maritime interference factors.
NETL
NSTL
万方数据
