Article: Real-time capture and simulation system of gymnastics motion based on 3D vision transformer and multi-objective optimisation Journal: International Journal of Information and Communication Technology (IJICT) 2026 Vol.27 No.93 pp.46 - 63 Abstract: Existing gymnastics motion capture and simulation methods struggle with high-dynamic movements, rapid joint changes, and unnatural simulations. To address these pain points, we propose an end-to-end intelligent gymnastics movement analysis framework integrating deep learning spatiotemporal representation with 3D deformable transformers. We introduce a 3D deformable attention mechanism for limb tracking and a modal feature fusion strategy combining RGB and bone pose data. Additionally, a multi-objective optimisation framework with heuristic rules and parallel computing improves simulation authenticity and efficiency. Experimental results demonstrate high accuracy: the average 3D coordinate error for difficult churning is 35.6 mm, key pose recognition is 78%, and swivel tracking accuracy reaches 91.2%. The real-time tracking delay is as low as 12.3 ms, significantly outperforming traditional methods. Inderscience Publishers - linking academia, business and industry through research

Title: Real-time capture and simulation system of gymnastics motion based on 3D vision transformer and multi-objective optimisation

Authors: Qichang Xu

Addresses: Institute of Physical Education, Hechi University, Yizhou, 546300, China

Abstract: Existing gymnastics motion capture and simulation methods struggle with high-dynamic movements, rapid joint changes, and unnatural simulations. To address these pain points, we propose an end-to-end intelligent gymnastics movement analysis framework integrating deep learning spatiotemporal representation with 3D deformable transformers. We introduce a 3D deformable attention mechanism for limb tracking and a modal feature fusion strategy combining RGB and bone pose data. Additionally, a multi-objective optimisation framework with heuristic rules and parallel computing improves simulation authenticity and efficiency. Experimental results demonstrate high accuracy: the average 3D coordinate error for difficult churning is 35.6 mm, key pose recognition is 78%, and swivel tracking accuracy reaches 91.2%. The real-time tracking delay is as low as 12.3 ms, significantly outperforming traditional methods.

Keywords: 3D vision transformer; multi-objective optimisation; gymnastics motion capture; action simulation; real-time performance.

DOI: 10.1504/IJICT.2026.155871

International Journal of Information and Communication Technology, 2026 Vol.27 No.93, pp.46 - 63

Received: 06 Apr 2026
Accepted: 11 May 2026
Published online: 24 Aug 2026 *

Title: Real-time capture and simulation system of gymnastics motion based on 3D vision transformer and multi-objective optimisation

Keep up-to-date