Research on underwater target tracking method combining deep learning and kernel correlation filtering

Detection and tracking of underwater targets is a prerequisite for autonomous operations of underwater robots. The optical imaging is often used for close range environmental perception for underwater robots because of its high resolution. Deep learning is widely used in object detection tasks, but it takes a long time and is rarely applied in tracking tasks with high real-time requirements. Since correlation filters have been applied to target tracking, their ultra-high speeds have attracted widespread attention. In this paper, a deep learning target detection method is combined with a kernel correlation filter to solve the real-time problem of underwater target tracking. In this paper, experiments are conducted using the fish image dataset Labelled Fishes in the Wild and the results show that the method in this paper is feasible.