Quantitatively scoring behavior from video-recorded, long-lasting fish trajectories

Abstract

Scoring animal behavior is increasingly needed for better understanding ecological processes. For example, behavior shapes harvesting likelihood, thus management of harvested resources should improve after accounting for behavior-driven processes. Automatic video-recording at controlled arenas is the most widespread method for scoring behavior. However, long term tracking animals while keeping identity is still an opened challenge. Here, we develop an ad-hoc algorithm for multi-tracking objects during days or even weeks, to fulfill the particular needs for a behavioral assay concerning a fish species targeted by recreational fishing. Specifically, we overcome the challenge of keeping fish identity in a context where they often disappeared from the camera when entering a shelter, the pixel size was low compared to the size of the arena and the lighting was constrained by the wellbeing of the fish. This work may contribute to better assess the behavioral features of fish in long-lasting lab conditions.