Research On Object Detection And Tracking Methods For USV Based On Monocular Vision

We have a vast maritime territory,further promoting the maintenance for our maritime rights and supervision of enforcement for the law at the sea in an all-round way is a solid and necessary support to our goal of building China into a maritime power.With the development of research and application of unmanned intelligence system,research on unmanned intelligence equipment for maritime has been a hotspot in the field of maritime technology.With high safety,high efficiency and autonomy,unmanned surface vehicle(USV)which integrates kinds of sensors has a great potential and worth for the applications in maritime rights maintenance,law enforcement and maritime rescue.Approaching monitoring and tracking play an import role in our maintenance of marine rights,and it relies on large official vessel for now,which comes with a high cost ang low efficiency.With the support of our USV and a mono-camera,aiming at the task of target ship monitoring and considering features of visual images and motion features of USV and target ship,the research of reliable detection and robust tracking methods for ship in the maritime is taken.For the maritime object detection,a novel detection method is proposed.First,the improvement on the FT(Frequency-Tuned)saliency detection is conducted,which can maintain the most saliency region in real time,and the saliency map is segmented by maximum saliency.Then,sea sky line is extracted by the Hough transform,and cloud region above sea sky line can be excluded.Last,dynamic disturbs from wave and wake region are inhibited by matching centroids in random chosen history frames.Therefore,maritime object can be detected accurately in real time.For the maritime tracking,considering scale variation and translation variation object,the fDSST(fast Discriminative Scale Space Tracker)method is adopted to estimate the position and size jointly.And the heading,pitch and roll of USV are used to change the learning rate of correlation filter and to estimate size and position of target ship adaptively.In the experimental verification section,the self-development USV system and the different designed tracking situations are introduced.The fusion framework of object detecting and object tracking is proposed,and four different encounter scenes are designed and taken.The real time and robust ability as well as efficiency are illustrated in the final visual tracking results.