Bear-vision: Using Artificial Intelligence to Detect Polar Bears

Climate change is driving rapid sea ice loss in the Arctic, threatening the habitat polar bears rely on for survival. As sea ice seasons shorten, polar bears are spending more time on land, increasing the risk of encounters with people. To help mitigate human-polar bear conflicts, the conservation organisation Polar Bears International (PBI) is developing an AI-powered early warning system called “Bear-vision”. Schönbrunn Zoo, which has supported PBI for many years, is contributing to this research project by providing video footage of its male polar bear.

Objectives

The aim of the research project is to develop a cost-effective early warning system that uses machine learning to automatically detect polar bears in video footage and provide early alerts when bears approach human activity. 

Relevance

Early detection can help protect both people and polar bears by reducing the risk of dangerous encounters and unnecessary defensive actions against bears. Polar Bears International has already developed and successfully deployed its radar-based detection system “Bear-dar” in the Canadian Arctic. As this technology is expensive, “Bear-vision” aims to provide a more affordable alternative. Scientifically managed zoos play a valuable role in this research, as obtaining video footage of wild polar bears in a wide variety of settings and environmental conditions is challenging. 

Methods

Schönbrunn Zoo is collecting video footage of its male polar bear under different weather and lighting conditions. Recordings taken in snowy and icy conditions, darkness and other situations with limited visibility are particularly valuable. Together with footage from other zoos and aquariums, these recordings will contribute to a comprehensive training dataset. The dataset will be used to evaluate and improve existing machine learning models, enabling automated detection of polar bears in different environments, lighting conditions and at varying distances.

Ongoing Research

The research project is expected to continue until 2028.