Research on the Statistical Method of Ship Flow based on Deep Learning and Virtual Detection Line

Abstract

In order to solve the problem of ship flow statistics in inland waterway, this paper proposed a tracking counting method which based on the deep learning single-target tracking network model and virtual detection line in ship video. Firstly, we extracted the ship objects by combining background difference and Otsu method, and then calculated the ship position information by using the projection method. Then, initialized the improved single-target tracking network model. Finally, we delineated the virtual detection line, and calculated the distance between the virtual detection line and the ship target output by the network model to realized the ship flow statistics. The experimental results show that the proposed method could achieve a ship traffic flow accuracy rate of 93.9% under the actual application scenario. It can be combined with the existing VTS system and has certain application value.