Automating camera trap labeling and deriving population densities using distance images

In 2022, I became the project manager for a new cooperative agreement with US Fish and Wildlife to help clear a backlog of unsorted camera trap images from sites throughout the southwest region. I built a participatory science project using the Zooniverse platform to collect labels for over a dozen common species found at wildlife management areas in New Mexico, Texas, and Oklahoma. I am currently working to create species-specific deep learning models for individual wildlife refuges, leveraging the popular MegaDetector model for pretraining. Our priority sites are Sevilleta National Wildlife Refuge in New Mexico and Laguna Atascosa National Wildlife Refuge in Texas, though we also plan to expand to Bosque del Apache NWR and Red Rocks Wildlife Management Area in New Mexico as well as Wichita Mountains NWR in Oklahoma.

I have also been assisting other collaborators on training preliminary deep learning models and deriving distance images using photogrammetry for the purpose of ungulate population density estimation at Sevilleta NWR.