AI driven automated behavior analysis

The observation of animals is one of the most important processes for understanding and conserving the diverse and complex animal world, regardless of whether they live in a controlled environment or in the wild. When animals are kept in human care, careful observation is essential to assess and improve animal health and behavior. Zoos and wildlife sanctuaries conduct systematic observations to monitor physical well-being, identify reproductive patterns and understand the social dynamics of captive populations. In the wild, animal observation is a cornerstone of ecological research and conservation efforts. Field observations provide invaluable insights into population sizes, species interactions, migration patterns, and habitat preferences. These observations form the basis for developing conservation efforts, identifying threats, and mitigating human impacts on natural ecosystems. Long-term studies of wildlife also provide important data for understanding the effects of climate change, habitat loss, species decline and other environmental problems.

Tom Burger - TGFN

Our Contribution

Animal observation carried out by biologists, ecologists and veterinarians can be divided into three steps: the localization of the animals in their habitat, the identification of individual animals, and the recognition of behavior. Especially the second step is crucial because it enables biologists and animal keepers to tailor possible measures or treatments to individual needs. In close collaboration, biologists and computer scientists have automated each of these steps. The AI-driven algorithms we have developed enable continuous monitoring of individual behavior. We can determine activity/inactivity patterns, stereotypical behavior, group interactions, the use of enclosures and distances traveled, among other things.It is also possible to train the AI models on very complex behavioral ethograms. For polar bears (Ursus maritimus), our AI models are able to distinguish over 170 different behaviors. The methods we propose are species-agnostic and can be transferred to other animal species and environments. In addition to the countless applications in zoological institutions, the methods we have developed are also relevant for research in the field. We apply our models to video data of wild animals to track individuals and their behavior over time. Among other projects, we support the Bavarian Forest National Park in evaluating photo traps to automatically monitor the lynx population.