A new artificial intelligence system discussed in the International Journal of Simulation and Process Modelling could help make school and university sports safer and more effective. The system can analyse video of human movements in real time and combine this with data from sensors to understand those movements. The analysis can identify movements that would be unclear to a human observer because they are fast, unpredictable or hidden from view. The researchers suggest it will be a boon in physical education and how it is taught and monitored.
Current “pose recognition” tools use computer vision to identify human postures but often struggle with the complexities of real-world sports. Movements are usually irregular and constantly changing in most sports. The new system tackles this underlying problem by combining visual information with data from motion sensors, and then building on embedded simulations to allow it to interpret and predict movements as they happen.
The team tested their system on established benchmark datasets, including Human3.6M and the team’s own CollegeSports-200 collection. They achieved almost 97 per cent accuracy. They kept the error count caused by visual obstructions low and with a frame rate of 38 frames per second, the system is fast enough to deliver live feedback. In field trials, they also had encouraging results as students were guided based on the system’s analysis and were seen to improve on poor posture and results when their teachers had a clearer picture of fitness patterns across groups.
The team suggests this novel approach could lead to more adaptive, evidence-based physical education. Such an approach uses data directly to plan improved exercise and practice, and even to cut the potential for injuries.
Chen, D., Ni, Z. and Huang, W. (2025) ‘Multimodal pose estimation and simulation modelling for real-time human motion analysis’, Int. J. Simulation and Process Modelling, Vol. 22, No. 5, pp.1–10.
No comments:
Post a Comment