Research on Real-Time Motion Capture System Based on Deep Learning

DOI: https://doi.org/10.62517/jike.202504420

Author(s)

Yiran Bo

Affiliation(s)

School of Information Science and Engineering, Northeastern University, Shenyang, Liaoning, China *Corresponding Author

Abstract

This study addresses the limitations of traditional motion capture systems, such as high cost, insufficient robustness, and poor real-time performance. It aims to develop a low-cost, highly robust real-time motion capture prototype system. The system is designed to achieve a joint position estimation error of no more than 5cm under occlusion and complex lighting conditions, with a delay of ≤50 ms to meet real-time interaction needs, and to verify 1-2 application scenarios like virtual human animation generation or gait analysis. Research methods include multi-angle data collection and preprocessing, adoption of a model framework integrating two-stream graph convolution, adversarial learning and trajectory space, as well as model compression (pruning, quantization) and hardware acceleration (GPU, TPU) to improve real-time performance. The system advances cross-disciplinary research on multimodal perception and human motion modeling, and provides technical support for film/animation production, VR/AR, sports analysis and medical rehabilitation.

Keywords

Real-Time Motion Capture; Deep Learning; Two-Stream Graph Convolution

References

[1] Richardson R T, Russo S A, Chafetz R S ,et al.Reachable workspace with real-time motion capture feedback to quantify upper extremity function: A study on children with brachial plexus birth injury [J]. Journal of Biomechanics, 2022, 132:110939 [2] Chemli Y ,M. Tétrault, Marin T ,et al.Super-resolution in brain positron emission tomography using a real-time motion capture system [J]. NeuroImage, 2023, 272:120056-120056. [3] Yi X, Zhou Y, Habermann M, et al. EgoLocate: Real-time Motion Capture, Localization, and Mapping with Sparse Body-mounted Sensors [J]. ArXiv, 2023, abs/2305.01599. [4] Pan S, Ma Q, Yi X,et al. Fusing Monocular Images and Sparse IMU Signals for Real-time Human Motion Capture [J]. 2023. [5] Rojik A, Khoury J. Real-Time Teleoperation of a Robot Arm for Self-Contact Bc[J]. 2023. [6] Huang H, Zhao L, Wu Y. An IoT and machine learning enhanced framework for real-time digital human modeling and motion simulation [J]. Computer communications, 2023(Dec.):212. [7] Lugris U, Perez-Soto M, Michaud F C J .Human motion capture, reconstruction, and musculoskeletal analysis in real time [J]. Multibody system dynamics, 2024, 60(1):3-25. [8] Shan W, Lu H, Jia C, et al. Real-Time Human Motion Transfer System for Holographic Displays [J]. 2024 IEEE International Conference on Multimedia and Expo Workshops (ICMEW), 2024:1-2. [9] Zhang T . Real-Time Sports Image Recognition System Based on Deep Learning Algorithm[C]//2025:324-331.