Long-term Robust Tracking For Ships Research On Key Technologies

With the rapid development of the shipping industry,the throughput of ships in key sections of the inland river and the number of ships in the sea area are increasing year by year.Safety accidents mainly caused by ship collision and piracy are increasing,which poses a huge threat to the safety of the shipping industry.Referring to the relevant literature,researchers have conducted extensive research on the detection and tracking technology of inland river monitoring system,but there is a lack of research on the problem of ship multi-classification tasks and the stern tug interference of moving ships.In view of the above problems,this paper studies the key technologies of ship long-term robust tracking combined with ship intelligent vision system.Ship intelligent vision system detects and classifies ships in the field of vision,which helps to find suspicious ships in advance and put forward early warning.Aiming at ship detection and classification technology,this paper proposes a ship detection and classification technology based on YOLOv3 to realize real-time monitoring of shipping ships.This algorithm combines YOLOv3 algorithm with dark channel prior dehazing algorithm to ensure that the detection algorithm extracts accurate deep semantic information in video frames and improves the accuracy of detection and classification.The experimental results show that this method can detect and classify ships in real time,and the detection accuracy reaches 89 %,which provides guarantee for the long-term tracking of subsequent ships.Ship tracking is a key link in the process of ship route prediction and counterattacking pirate ship robbery.Aiming at the problem of stern towing interference and target occlusion in tracking,a long-term ship tracking method is proposed to suppress stern towing,aiming at improving ship tracking accuracy and realizing long-term tracking.First of all,image preprocessing to enhance image clarity;Then in the tracking,the off-line trained tug detector is used to detect the tug position on the water surface,and the ship tracking results are corrected in real time;In the tracking,the target occlusion is judged according to the perceptual hash algorithm.When the occlusion occurs,the online detector is used to detect the target area,and the ship position is recovered to initialize the tracker to realize long-term tracking.The experimental results show that the method can effectively suppress the influence of stern tug on tracking,improve the accuracy of ship tracking results,and realize long-term ship tracking.