SUBTLE can study and categorize the intricate movements of animals. (Image courtesy of the Institute for Basic Science)
DAEJEON, May 22 (Korea Bizwire) – In a breakthrough for animal behavior analysis, researchers from the Institute for Basic Science (IBS) in South Korea have developed an artificial intelligence (AI)-powered tool called SUBTLE that can study and categorize the intricate movements of animals.
The ability to analyze animal behavior is crucial for various research fields, including basic neuroscience studies, uncovering the causes of diseases, and exploring potential treatments.
While AI-based techniques have recently gained traction for understanding the complex, time-varying behaviors of animals, existing methods relying on single-camera recordings and manual labeling of specific movements have proven time-consuming and susceptible to human bias.
The research team, led by Jeon Chang-Jun and Cha Meeyoung, adopted an “unsupervised learning” approach. By analyzing motion data extracted from 3D motion capture equipment, they successfully implemented a framework that can classify animal behaviors without the need for labeled training data.
Unsupervised learning allows AI systems to identify patterns, structures, and characteristics within unlabeled data, grouping similar instances based on their inherent similarities. This minimizes human intervention and potential biases in analyzing animal behavior.
SUBTLE can study and categorize the intricate movements of animals. (Image courtesy of the Institute for Basic Science)
When tested on mouse movements, SUBTLE accurately distinguished various behavior patterns, such as rearing on hind legs, walking on all fours, pausing, and grooming, achieving over twice the accuracy of existing analysis methods.
The tool’s capabilities extend beyond mouse behavior, enabling comparisons between groups like juvenile and adult mice, and precisely identifying human exercise movements like push-ups, lifting, and pushing.
In April of last year, the research team transferred the SUBTLE technology to ActNova, an AI-based clinical and preclinical behavior testing analysis company, and has made the tool’s code available as open-source software.
“The algorithm developed by our team can be widely applied not only in robotics industries that mimic animal movements but also in any industry that requires behavior pattern recognition,” said Cha, the lead researcher.
The findings of this study were published in the latest issue of the International Journal of Computer Vision on May 20.
Kevin Lee (kevinlee@koreabizwire.com)
