Real-Time Tracking and Positioning of Underwater Visual Targets Based on Siamese Network

In this paper, the real-time tracking and positioning problem of underwater targets under the monocular vision system is investigated based on Siamese network. First, to solve the problem that monocular vision system cannot spatial positioning, an artificial marker with multi-features information is designed. On this basis, the tracking module and detection module are respectively designed to achieve real-time stable tracking and precise positioning of underwater targets. The experimental results of three sets of different environments show that the detection rate can reach 20FPS, and the positioning error does not exceed 1.4 cm, which demonstrates the real-time performance and strong generalization ability of the method proposed in this paper.