Researchers have developed an AI-based system that can track the behavior of brown bears in the wild and identify whether a DNA sample is likely to be found at a particular location.
Bear tracking with AI
AI Bear Tracking: Monitoring bear activity and collecting DNA samples could become easier than ever. Researchers have developed an AI-based system that can track the behaviour of brown bears in the wild and identify whether a location is likely to yield DNA samples. It is believed that this technology could prove to be a major aid in wildlife conservation and research.
This research, published by the Zoological Society of London, explains that AI can be used to analyse images captured by camera traps to identify specific bear movements. Specifically, the system monitors when bears stand on two legs, as this behaviour is often seen when they rub their bodies against trees.
How are bear DNA samples collected?
Wildlife experts have long been collecting DNA samples from bear fur. Barbed wires, called hair snares, are attached to trees to do this. When bears rub against the tree, some of their hair gets caught in the wires, which are then used for DNA testing. However, this process requires researchers to repeatedly visit these locations to collect samples, which consumes significant time and resources. Furthermore, even inspections often fail to yield useful samples.
How did AI recognize the behavior of bears?
The researchers used 2,373 camera trap images collected between 2020 and 2023. Since no dataset on bear activity was available, the scientists manually identified 15 key body parts in the images. A YOLOv11-based pose estimation model was then trained. This model proved to be 93.2 percent accurate in identifying bear body postures. A separate machine learning model classified bear postures, such as bipedal and quadrupedal, with 96.1 percent accuracy.
Will work even in remote areas
According to the researchers, this system is also useful in areas where mobile network or internet connectivity is limited. While camera-type images typically require significant data transmission, the AI system can transfer only essential pose data and simple visual information via satellite link instead of sending the entire image. This reduces the data transfer size from several megabytes to approximately 100 bytes, making surveillance even in remote areas easier.
