This work is a collaboration with A. Morand from the CEREMA group “Biodiversité Aménagement et Nature en Ville”, with contributions from the association “Les Amis de L’Erdre”, and GT-Europe students conducting research projects. The core idea stems from the need to monitor wild-life underground tunnels dedicated to the preservation toads, frogs, newts and salamanders. Because these animals don’t trigger InfraRed motion detectors, monitoring cameras are taking images every few seconds, resulting in thousands of images per night and per site. Monitoring multiple site over several months, easily results in hundreds of thousand images.

This is where machine learning comes into play. Using YoLo detectors, one can train a deep learning framework to detect and classify individuals, as illustrated below.

Toad detection principle

A prototype version of a toad tracker estimating the trajectories of the toads and counting them can be seen in the video below

To make for less blurry images taken by the Reconyx Hyperfire classes at short range, we designed a lens holder, which is available here in open-access: https://github.com/cedricpradalier/reconyx-glasses.