Minimizing Human Labor for In-the-Wild Camera Trap Processing Pipeline
Haoyu Chen, Amy R. Reibman · 2024
Camera traps are an important tool in ecological studies for non-intrusive monitoring of various animals. However, annotating camera trap data usually requires large amount of human labor. Therefore, we propose a solution for practitioners with limited human resources: an automated pipeline for curating in-the-wild camera trap data that mimics human annotators and significantly reduces the amount of human labor needed. We also propose evaluation protocols for estimating system performance on unlabeled data, and present experiments that demonstrate our pipeline's strengths, weaknesses, and flexibility to accommodate users' requirements.