Virtual reality (VR) has steadily become a key tool in sports training, offering immersive environments that simulate real-world physical exercises. However, its application in aerobics training has faced significant challenges, particularly in accurately capturing and recognizing complex body movements. A study in the International Journal of Computational Systems Engineering has overcome some of these barriers obstacles by improving the precision of motion capture and the recognition of different actions during aerobics exercises.
Conventional VR motion capture systems rely on algorithms that align 3D representations of physical objects, so-called point cloud data, to track body movements. However, these systems often struggle with two critical issues: noise and incomplete data. Noise refers to unwanted interference that can distort the data collected by sensors, while incomplete data arises when certain body movements are not fully captured. In aerobics, where precision in movement is key to safe and effective exercising, these issues compromise the effectiveness of VR-based training systems.
Hui Wang of the School of Physical Education at Yan’an University in Yan’an, China, has addressed these challenges by developing two new models. The first focuses on enhancing the Iterative Closest Point (ICP) algorithm, which is used to align point cloud data. ICP is a well-established method, but it is prone to inefficiencies, particularly when faced with noisy or incomplete data. By optimizing the algorithm, Wang has improved accuracy and speed of capture.
The second model focuses on refining action recognition. A neural network is used to analyse the complex relationships between body joints over time by tracking the interactions between different body parts. Wang improved the neural network used by incorporating a perturbation mechanism to deal with noise, which further improved its ability to capture subtle movements and interdependencies between non-adjacent joints during aerobics.
Accuracy up to 99 percent was achieved, indicating a remarkable ability to recognize and classify aerobics movements with minimal error. Moreover, the experimental group using these advanced models outperformed the control group in various performance metrics, particularly in terms of the standardization of movements, the work explains. This enhanced motion recognition technology could significantly improve both the learning experience for students and the ability of instructors to offer targeted feedback, leading to more efficient and personalized training in aerobics.
Wang, H. (2025) ‘The application of VR-based fine motion capture algorithm in college aerobics training’, Int. J. Computational Systems Engineering, Vol. 9, No. 6, pp.1–10.
