An international team, including researchers from the University of Bristol, has developed an AI system called SA-FARI. This system automatically finds, names, and follows individual animals in video footage, even tracking the exact outline of a specific bird as it flies through a dense forest.
Scientists have amassed billions of animal location data points over years of research. However, the capacity to automatically track and analyze the precise, individual behaviors of these animals within video footage has remained largely out of reach. This gap meant much of the granular detail about daily animal lives stayed hidden.
This new system suggests conservation efforts are poised to become significantly more targeted and effective. Yet, such granular insight also introduces new ethical considerations for animal monitoring and data privacy, requiring careful implementation.
How AI Studies Animal Behavior
The ecological community already collects vast amounts of data on animal movements. Scientists work to understand species distribution and migration patterns across various habitats.
- The Movebank database, maintained by the Max Planck Institute, has logged nearly 11 billion animal locations spanning over 1,600 species as of April, according to Nature.
- Ultralight GPS tags can now broadcast an animal's location every second, providing much more detailed behavioral data than previously possible, according to Nature.
Despite these advancements, even high-frequency location data fundamentally falls short of providing the visual behavioral analysis that SA-FARI offers. For years, conservation efforts often operated with an incomplete picture, mistaking 'where' an animal was for 'what' it was actually doing. This limitation meant crucial behavioral nuances, vital for understanding species survival, remained largely unobserved.
SA-FARI's Breakthrough
SA-FARI achieves its precise tracking through a specific technical innovation: 'masklets'. These 'masklets' represent the exact outline of an animal in a video from frame-to-frame, allowing for individual and behavioral analysis, according to the University of Bristol.
The AI system underwent extensive training, benchmarked on a dataset of over 11,000 wildlife videos from natural habitats. This training allows it to detect, name, and track animals of approximately 100 species, as reported by the University of Bristol. This 'masklet' technology, honed on extensive datasets, provides an unprecedented level of detail for individual animal identification and behavioral analysis, far beyond simple location data.
If SA-FARI's capabilities are widely adopted, ecological research will likely shift from broad population movements to granular individual actions, revealing previously hidden drivers of species survival and decline in 2026.
