Monitoring the Mobility of Sea Turtles in Video Footage

Monitoring the behavior of marine fauna in captivity is an efficient method for understanding the impact of anthropogenic activities as well as exploring behavioral indicators. Sea turtles, which are migratory animals, are particularly difficult to track in-situ and efforts to understand their physiology and behavior are mostly conducted in captivity.

A particular interest is to study the dose–response relationship of stress in sea turtles through monitoring their activity over time.

An efficient, non-invasive approach for this is behavioral analysis through the processing of video footage of sea turtles to detect changes in their dynamics.

However, video recordings produce large files that are impossible to process manually, and their analysis requires the application of automatic pattern recognition and classification techniques.

In this research, we present a classification model for monitoring the behavior of a single sea turtle in a rehabilitation water tank. Our method detects breathing events when the sea turtle raises its head above the water level, thus identifying excessive breathing which is an indicator of stress: the so-called fight-or-flight response.

Our processing is based on the ResNet model, pre-trained for image classification with training for the top two layers.

Training and testing was performed on video footage from two cameras overlooking the sea turtles.

Through manual tagging, we identified 500 instances where the turtle’s head was above the water level.

The results show an accurate and robust detection ability.

The model will be used in a research project to test the stress response of sea turtles to shipping noise disturbance.