Fish Tracking Based on Feature Fusion and Scale Adaptation in a Real-World Underwater Environment

Abstract Fish tracking is an important component of analyzing fish behavior and estimating fish population density. Due to the high degree of freedom of fish motion as well as the complex natural underwater environment, most existing object tracking methods are not ideal for fish tracking. In this paper, a fish tracking method based on feature fusion and scale adaptation is proposed, which is built on a kernelized correlation filter (KCF) to achieve accurate and rapid tracking. The proposed method mainly focuses on feature selection and scale estimation in the KCF framework. In feature selection, the color-naming feature and the histogram of oriented gradients feature are fused to improve the fish appearance model and reduce the influence of the high degree of freedom of fish motion and the complex natural underwater environment. In the scale estimation, an adaptive scale estimation scheme is employed to adapt the fish scale variation by learning a 1-D scale correlation filter. The experimental results show that the proposed method is effective and accurate for fish tracking in real-world underwater environments.