Professor Lee Young-moon's team, Hanyang University Department of Robot Engineering,  recently developed ‘PosePlusSeg’, which recognizes and analyzes human poses, according to Hanyang University on January 28. PosePlusSeg is a technology that allows AI to recognize people and estimate individual behavior at the same time. The technology is expected to be used in motion capture as well as in various AI fields such as applications in which AI recognizes multiple people simultaneously and provides customized services for each individual in the future.

In order for AI to recognize human behavior and interact optimally according to the situation, technology to accurately recognize people and grasp human behavior is needed. However, with the previous technology, the recognition's accurateness varies depending on the person's pose, and when the number of people increases in a specific area, the analysis speed slows down since the number of computations increases as well, which decreases utilization eventually.

To solve the problem, Professor Lee's team used a bottom-up method to find the body part first and estimate the pose, instead of the previous top-down method of finding the person first and then finding the body parts to estimate poses.

The key of PosePlusSeg is to estimate people's perceptions and poses without increasing computational volume through multi-task learning, thus improving overall performance. It also learn by recognizing pixel-style objects and estimating poses at the same time, not the previous way(bounding box; a method of explaining where an object is located in an image of a rectangle).

As a result, PosePlusSeg significantly improved the speed to more than five times faster(28 fps) than Facebook's Mask R-CNN(a computer vision algorithm that detects and identifies objects' peripheral boundaries), which enables real-time recognition and analysis of AI.

PosePlusSeg will be announced at the "AAAI Conference on Artificial Intelligence(AAAI)" held from February 22 to March 1 in recognition of its originality and excellence. AAAI is the flagship conference of the Association for the Advance of Artificial Intelligence and is known for the best conference in the field of artificial intelligence recognized worldwide.

Meanwhile, the study was conducted with the support of the Artificial Convergence Research Center, the Korea Research Foundation, and the Information and Communication Planning Agency, with Niaz Ahmad, Jawad Khan, researcher Kim Yoo-hyun, and Professor Lee Young-moon participating.