3-D Video Tracking of Multiple Fish in a Water Tank

Abstract

Study on fish behavior is essential to fishery development and water environment protection. Quantitative analysis of fish behavior is impossible without information on fish trajectories. Although the computer vision techniques have provided an effective approach to the collection of fish trajectories, it is still challenging to track fish groups accurately and robustly due to fish appearance variations and frequent occlusions. In this paper, a skeleton-based method for 3-D tracking of fish group is proposed. First, skeleton analysis is performed to simplify the top- and side-view objects into representation of feature points. Next, the feature points under the top view are associated to obtain the top-view trajectories of objects. Finally, the epipolar constraint and the motion continuity constraint are used to match the top-view trajectories with the side-view feature points, thereby obtaining the 3-D trajectories of objects. Experimental results demonstrate the ability of the proposed method to track fish group more effectively than other state-of-the-art methods.