Dynamic frame-based weight estimation and joint position-scale model for underwater object tracking

Due to the complex underwater environment and the scattering and absorption effects in water, underwater images often suffer from low visibility and weak texture features, posing significant challenges for object feature extraction and tracking. To address these issues, we propose a novel underwater object tracking method based on dynamic frame selection and a joint position-scale estimation model. Our approach initially employs a dichotomy method to select the most correlated frame with the current frame, using inter-frame information to estimate the weights of convolutional features for effective feature fusion. Subsequently, a tracking model with joint position-scale estimation is constructed, where the fused object features are input to estimate the object’s position and scale accurately in complex underwater environments. Additionally, a confidence evaluation metric based on average peak correlation energy is integrated into the model update strategy to halt updates during object occlusion or disappearance, enhancing tracking stability and preventing error accumulation. Experimental results on the UTB180 dataset and actual underwater environments demonstrate that our algorithm achieves improved tracking precision and success rates compared to state-of-the-art methods, particularly in scenarios with scale variations, occlusions, and low visibility.