Research On Ship Tracking Algorithm Based On Machine Vision For Harbor Hannel

The shipping in harbor channel is an important strategic resource for the sustainable development of China’s economy and the ports internationalization of "The Belt and Road".It plays an indispensable role in the entire transportation system.Due to some factors,such as investment and performance,traditional maritime patrol has been unable to meet the current requirements of port waterway regulation.So the dynamic ship monitoring based on visual technology has become the development trend of ship on-site monitoring.The real-time dynamic tracking of under navigation is one of the core elements of the shipping safety supervision system in the port area.Therefore the research of this topic not only has certain practicability and frontier,but also has certain promoting effect and potential market value for the development of related disciplines.(1)The design of port domain ship tracking data set and experimental cloud platform aims at the problem that there is no open standard database in port ship tracking.From the performance evaluation system in target tracking and the present situation of port domain ship dataset,the design of tracking data set and experimental cloud platform is introduced.Finally,the functional design and characteristics of the platform are described and analyzed.(2)In order to facilitate the study of region adaptive random projection tracking algorithm,this dissertation first studies visual tracking from open dataset.When only the initialization position of the target is obtained,it is a very challenging task to achieve robust tracking in continuous sequences.In order to alleviate the change of target appearance caused by illumination and occlusion interference,a target appearance model based on sub-region is designed in this dissertation.The model uses feature fusion to skillfully fuse the color,texture and spatial structure features of the target compression domain.In this dissertation,a novel random observation matrix is designed to construct an adaptive model to reduce tracking drift..At the same time,the confidence distribution of sub-region can effectively represent the target appearance change,and the median flow tracking module can stably estimate the scale change in adjacent frames,so that the tracking model can update the parameters of the classifier in a timely manner.A large number of experiments in visual tracking standard database and selfcollected ship data show that the algorithm is superior to some state-of-the-art algorithms in accuracy,robustness and so on.At the same time,through the experimental comparison of specific port domain scenarios,it provides positive and diverse enlightening information for other chapters to construct more effective port domain ship tracking algorithm.(3)The real-time multi-scale ship tracking algorithm in port channel has achieved favorable results in face recognition and human tracking of long-term unconstrained video stream by using predator algorithm.The predator algorithm has two obvious defects in ship monitoring in port channel CCTV system:(1)when there is occlusion between ships,shortterm tracking strategy is prone to tracking drift;(2)the cascade target detector can not adapt to the scale change of ship motion when the current video frame searches the interested moving ship.In this dissertation,the above problems in predator system are deeply analyzed,and the problem of ship tracking in port and waterway under occlusion and scale change is solved.In this dissertation,the compression coding theory is applied to the predator short-term tracking strategy to improve the ship tracking problem under occlusion.Secondly,an adaptive scale strategy is proposed to solve the scale change problem in the process of target motion.The experimental results show that the robustness of the proposed algorithm to occlusion and scale is obviously better than that of the original algorithm.(4)Research on context-aware correlation filtering ship tracking algorithm aims to solve the problem of long-term and serious occlusion.Combined with the traditional correlation filter framework and context awareness algorithm,a correlation filtering tracking method based on context awareness is proposed,which cooperates with sample selector and occlusion tracker.The experimental results show that the favorable accuracy,success,robustness to large serious occlusion,background clutter and scale change,and the stability of the proposed algorithm are significantly higher than those of other comparative algorithms,which fully proves the superiority of the algorithm.