Efficient Track Association of Ships through AIS and shipboard Video Fusion via Ad Hoc Networks

This paper presents a shipboard equipment-based method that fuses multi-camera and AIS sensing information, enabling intelligent sensing and recognition of encountered ships. This approach maximizes the strengths of both AIS and shipboard cameras, achieving more precise, comprehensive, and rapid sensing of surrounding vessels. To address issues with measurement accuracy and range in existing methods, a dual-camera ship-ranging technique is proposed for spatial localization using video image data. Utilizing multi-camera setups, ship targets are detected and tracked. The ship detection algorithm provides target frame information, which is then used in the dual-camera ranging method to convert ship pixel coordinates into the world coordinate system. To improve track correlation rates, a novel two-branch track correlation network, with an LSTM backbone network, is introduced, enabling ship track correlation through deep learning. By establishing an AIS and video track association model based on deep learning algorithms and creating a ship track association dataset, an efficient and reliable track association process is achieved.