Underwater Fish Tracking-by-Detection: An Adaptive Tracking Approach.

High distortion and complex marine environment pose severe challenges to underwater tracking. In this paper, we propose a simple, template-free Adaptive Euclidean Tracking (AET) approach for underwater fish tracking by regarding tracking as a specific case of instance detection. The proposed method exploits the advanced detection framework to track the fish in underwater imagery without any image enhancement techniques. The proposed method achieves comparable performance on the DeepFish dataset, with 22% and 14% improvement in precision and success over state-of-art trackers.