Representation of Trajectories for Fish Anomaly Detection

Develop a robust and generalizable representation of fish movement trajectories for trajectory-based anomaly detection in aquaculture monitoring systems, ensuring that the representation adequately captures the characteristics of fish motion necessary for detecting abnormal behaviours from tracked trajectories.

Background

The paper discusses visual monitoring systems that extract trajectories of multiple fish to model and detect abnormal behaviours. It notes that most trajectory-based anomaly detection approaches have relied on clustering or statistical modeling, and emphasizes that the way trajectories are represented is central to the success of these methods.

Within this context, the authors explicitly state that the representation of trajectories remains unresolved. They also highlight that fish trajectories inherently encode position, speed, and direction, and that defining abnormality may involve multiple aspects, underscoring the need for a suitable and comprehensive trajectory representation for reliable anomaly detection in aquaculture.

References

In the past decade, track-based anomaly detection methods have primarily relied on traditional clustering methods or focused on statistical models of trajectories. In contrast, the representation of trajectories remains an open problem .

— Fish Tracking, Counting, and Behaviour Analysis in Digital Aquaculture: A Comprehensive Survey  (2406.17800 - Cui et al., 2024) in Section: Fish school behaviour analysis based on Trajectory analysis