BatDetector: modular open source pipeline for bat monitoring: with Natagora

Details

Supervisors

Jodogne, Sébastien

Faculty

Ecole polytechnique de Louvain

Degree label

Master [120] en sciences informatiques, à finalité spécialisée
Master [120] en sciences informatiques, à finalité spécialisée

Abstract

enWe present a modular open source two‐stage computer vision pipeline for automated bat detection, developed in collaboration with Natagora to facilitate large scale monitoring of roosting behavior. Natagora’s field sites provide a unique opportunity to capture bat activity and understanding roost locations is critical for conservation so that they can recommend and construct new shelters that support local biodiversity. However, manual annotation of bat images is exceedingly time‐consuming, often requiring experts to review thousands of frames where bats occupy only a small proportion of the picture, a small‐object detection challenge. To address this, our pipeline first performs background subtraction to isolate candidate foreground blobs. These blobs are then passed to a binary classifier that discriminates true bat instances from background artifacts. This modular two‐stage approach drastically reduces the annotation burden in order to reallocate precious time toward more productive ecological analysis.