Researchers have developed an artificial intelligence (AI)-based system designed to assist wildlife experts in tracking brown bear behavior and enhancing DNA collection efforts in remote areas. The study, published by the Zoological Society of London, details a method that leverages camera-trap images and AI models to identify when brown bears stand on two legs near hair snares—a behavior often associated with rubbing against trees and leaving fur samples suitable for DNA analysis.
Challenges with Current Hair Snare Methods
Hair snares, which consist of strands of barbed wire attached to trees, are commonly employed to collect bear fur without disturbing the animals. However, researchers currently must visit these sites regularly to gather samples and prevent contamination, a process that demands significant time and resources. According to the study, only about 24% of inspections at hair snare stations in the Catalan Pyrenees yield usable samples, underscoring the need for a more efficient monitoring system.
How the Proposed AI System Works
The researchers trained an AI model to detect whether a bear is standing on two legs (bipedalism) or moving on four legs (quadrupedalism) using images captured by camera traps. Since no annotated bear pose dataset existed, the team manually labeled 2,373 images collected between 2020 and 2023, marking 15 anatomical points on each bear's body. Using the YOLOv11 pose estimation model, the system achieved a keypoint detection precision of 93.2%. A separate multilayer perceptron (MLP) model then classified bear postures with 96.1% accuracy, helping to identify potential interactions with hair snares. Researchers noted that bipedal behavior can indicate a bear has rubbed against a tree-mounted snare and likely left behind fur suitable for DNA testing.
Reducing Field Visits in Remote Areas
The proposed system is designed to operate in regions with limited or no cellular connectivity. Instead of transmitting full-resolution images, the AI can send lightweight pose data and simplified visual representations via satellite links, reducing data transfer requirements from several megabytes to about 100 bytes. The researchers believe this approach could help conservation teams focus field inspections on locations where bear activity has actually been detected, reducing unnecessary visits and improving the efficiency of wildlife monitoring programs.
Broader Implications for Ecological Monitoring
The study concludes that AI-powered pose estimation can be adapted for ecological monitoring tasks and may help automate parts of the DNA collection process while providing additional information about bear behavior that traditional hair snares alone cannot capture. This innovation represents a significant step forward in wildlife conservation technology, offering a scalable solution for monitoring elusive species in challenging environments.
