Ship tracking for maritime traffic management via a data quality control supported framework

Ship trajectory in maritime surveillance videos provides crucial on-site traffic information (e.g., ship speed, traffic volume, density) to help maritime traffic situation awareness and management in the smart ship era. To that aim, many focuses are paid to track ships from maritime videos by exploring distinct visual features from maritime images, which may fail under complex maritime environment interference (occlusion, sea clutter interference, etc.). The study proposes a novel video-based ship tracking framework with the help of Multi-view learning model and data quality control procedure. First, we obtain raw ship positions from maritime images with particle filter and Multi-view learning models. Then, a data quality control procedure is implemented to suppress ship tracking outliers with the help of Kalman filter. Finally, we verify our proposed model performance on three typical maritime traffic situations (ship occlusion, sea clutter interference and small ship tracking).