Real-time multimodal fusion for occlusion-robust vessel tracking in electronic imaging systems

Real-time and precise vessel tracking in inland waterway surveillance videos is essential for ensuring navigational safety and optimizing traffic management. To address tracking failures caused by vessel occlusions in complex scenarios, such as bridge areas, we propose a robust multimodal adaptive tracking framework tailored for inland waterway vessel monitoring. The framework employs the YOLOv11 object detection algorithm to extract vessel target information from video frames, complemented by a Kalman filter for state prediction and updating. A multimodal association model integrates kinematic features, hue-saturation-value appearance features, velocity vector features, and dynamic-time-warping-based trajectory matching to achieve robust target association. In addition, an adaptive predictive state transition mechanism significantly enhances trajectory tracking accuracy under occlusion conditions. Experimental evaluations in typical inland waterway occlusion scenarios demonstrate that the framework achieves a multiple object tracking accuracy of 0.89, an identity switch of 9.3%, and an occlusion success rate of 87.5% on surveillance video sequences.