Does Computer Vision Reliably Process Camera Trap Images in Multispecies Systems?

Abstract

Camera trap surveys are a useful tool to monitor wildlife populations and collect data to estimate various metrics, including population size. Camera surveys can generate large number of photos across various animal groups, but the time and effort it takes for researchers to process these datasets present a significant barrier to camera-based wildlife research. Artificial intelligence (AI) has emerged as a way to drastically speed up data processing time. However, any limitations to AI’s ability to accurately label images might have consequences for estimating population size. In this study, we compare species classification done by AI to a manual review. We use computer vision incorporated into Wildlife Insights to label camera trap images from species with morphological similarities in northwest Montana. The findings of this study help us determine if we can rely on available computer vision tools to classify images, and then use these data to reliably estimate population size.