WOE-VesselTrack: Robust Vessel Position Measurement Amidst Weather-Degraded Vision and Target Occlusion Scenarios for Bridge Collision Warning

Real-time monitoring of vessel navigation conditions in bridge waterways is essential for alerting to potential vessel-bridge collisions, which can be achieved by tracking a vessel’s trajectory in video, and fusing it with the complementary physical position information from the automatic identification system (AIS) to measure the position of the vessel. However, adverse weather conditions (e.g., foggy and rainy) would degrade video visibility, while occlusion could lead to losing the vessel’s visual features, causing inaccurate or failed tracking. To tackle these challenges, a novel monitoring method WOE-VesselTrack (Weather and Occlusion Enhanced Vessel Tracking) is proposed. It consists of three components: (1) The recent deep learning algorithm DeepSORVF is adopted as a pipeline to extract and fuse the video and AIS trajectory data of a vessel, realizing a basic tracking function. (2) A degradation-type guided image restoration (DGIR) model is constructed to improve the visual quality of weather-degraded video. DGIR learns to predict the residual between the degraded image and the clear image with the guidance of weather type by supervised training, achieving video quality improvement adaptively. (3) An adaptive triggering and Gaussian smooth interpolation (AT-GSI) algorithm is proposed to enhance the occluded trajectory prediction. It triggers the anti-occlusion operation adaptively based on the changes of detected vessel boxes, and predicts the occluded position by utilizing nonlinear fitting historical trajectories. The proposed WOE-VesselTrack is validated on the benchmark dataset FVessel. The results show notable tracking accuracy improvements in the metric MOTA, with an average increase of 17.4% by incorporating DGIR and 4.3% by integrating AT-GSI. These significantly reduce tracking failures and improve trajectory precision, providing reliable vessel status information for the vessel-bridge collision warning system.