Eyes in the Wild and Minds in the Loop: A Systematic Review of Artificial Intelligence for Camera-Trap Wildlife Monitoring

Camera traps generate imagery faster than people can review it, and artificial intelligence (AI) is increasingly used to analyse it. This systematic review asks to what extent AI applied to camera-trap imagery has moved from benchmark performance to ecologically usable monitoring. Following PRISMA 2020, Scopus was searched on 6 August 2026 for 2020–2026. Peer-reviewed English-language journal articles applying automated inference to imagery from unattended cameras observing free-ranging animals were included; captive or web-sourced imagery, non-image methods, reviews and conference papers were excluded. Of 637 records, 231 were read in full and 201 were included. Studies were appraised against six quality criteria, findings were synthesised as descriptive counts across five research questions, and performance was tabulated on each study’s own metric, without meta-analysis. Output grew by about 29% per year. Detection and species classification each appear in 62% of studies, behaviour recognition in 5%, and study regions invert global biodiversity gradients. Only 62% of studies evaluate beyond their training data, a share unchanged since 2020; the measured generalisation penalty ranges from six to nearly seventy points. Only 35% report field deployment; costs appear only in deployment-oriented studies. The gap between benchmark performance and ecological practice remains the field’s central challenge.