Skip to main content
ARS Home » Plains Area » Kerrville, Texas » Knipling-Bushland U.S. Livestock Insects Research Laboratory » Cattle Fever Tick Research Unit » Research » Publications at this Location » Publication #394104
Research Project: Integrated Pest Management of Cattle Fever Ticks

Location: Cattle Fever Tick Research Unit

Title: Automated tool to detect, classify, and count animals in camera trap images using artificial intelligence

Author
item TABAK, MICHAEL - Western Ecosystems Tech, Inc
item FALBEL, DANIEL - Western Ecosystems Tech, Inc
item HAMZEH, TESS - Western Ecosystems Tech, Inc
item BROOK, RYAN - University Of Saskatchewan
item Goolsby, John
item ZOROMSKI, LISA - Texas A&M University
item BOUGHTON, RAOUL - Mosaic Co
item SNOW, NATHAN - Animal And Plant Health Inspection Service (APHIS)
item VERCAUTEREB, KURT - Animal And Plant Health Inspection Service (APHIS)
item MILLER, RYAN - Animal And Plant Health Inspection Service (APHIS)

Submitted to: Trade Journal Publication
Publication Type: Abstract Only
Publication Acceptance Date: 4/15/2022
Publication Date: 2/7/2022
Citation: Tabak, M.A., Falbel, D., Hamzeh, T., Brook, R.K., Goolsby, J., Zoromski, L.D., Boughton, R.K., Snow, N.P., Vercautereb, K.C., Miller, R.S. 2022. Automated tool to detect, classify, and count animals in camera trap images using artificial intelligence. Trade Journal Publication. https://doi.org/10.1101/2022.02.07.479461.
DOI: https://doi.org/10.1101/2022.02.07.479461

Interpretive Summary: Motion activated game cameras are used to track their movements through fence crossings. Manual processing and classifying of the hundreds of thousands of images taken by the cameras can take many months to complete. The computer program validated in this research, can rapidly process images of animals and determine if the image contains a nilgai or other animals of interest. This program has the potential to increase the efficiency of wildlife studies, especially those involving wild pigs, nilgai and wildlife.

Technical Abstract: Motion-activated wildlife cameras, or camera traps, are widely used in biological monitoring of wildlife. Studies using camera traps amass large numbers of images and analyzing these images can be such a large burden that it limits the application of camera traps and in turn study design. We trained deep learning computer vision models using data for 168 species that automatically detect, count, and classify common North American domestic and wild species in camera trap images. The models performed well on both validation datasets and eight out-of-sample datasets. Out of sample wild pig recall accuracy was 99.1% for individual images and increased to 99.9% for camera trigger events containing three images. We provide our trained models in an R package, CameraTrapDetectoR. Three types of models are available: a taxonomic class model classifies objects as mammal (human and non-human) or avian; a taxonomic family model that recognizes 33 mammal, avian, and reptile families; and a species model which recognizes 77 domestic and wild species including all North American wild cat species, bear species and canid species; and each model also includes a category for vehicles and empty images. CameraTrapDetectoR can be used via a simple point-and-click graphical user interface. This tool can be used to minimize manual classification of camera trap images supporting the implementation of larger studies and near real time automated monitoring of populations.