The Efficiency of AI Motion Capture for Fighting Poses
Keywords:
AI Motion Capture, combat poses, animation accuracy, animation technologyAbstract
The use of Artificial Intelligence–based Motion Capture (AI Mocap) platforms requires importing original video clips recorded from various camera angles. A major challenge is the potential introduction of errors into the resulting motion data. This study aimed to examine the effectiveness of AI Mocap in capturing martial arts movements. The research employed a 3 × 3 factorial experimental design with two independent variables: (1) camera angle of the source video clips (front, diagonal, and side views) and (2) difficulty level of the martial arts movements (easy, moderate, and difficult). A total of nine distinct martial arts movements were included in the study. The dependent variable was the performance quality of the motion capture generated by the AI Mocap system. Data collection involved 27 animated clips derived from the captured motion data. These clips were evaluated by experts using an assessment form based on three criteria: accuracy, fluidity, and realism of the captured movements. The results revealed that: (1) The performance of AI Mocap technology in capturing martial arts movements across different levels of difficulty was consistently high, with no statistically significant differences observed (p > .05). Movements classified as easy demonstrated the highest performance in terms of motion fluidity. (2) The camera angle of the original video significantly affected motion capture performance. The front view provided the best results in terms of accuracy and realism, while the side view yielded the highest fluidity. In contrast, the diagonal angle produced significantly lower performance across all dimensions—accuracy, fluidity, and realism (p< .05). These findings indicate that camera angle is a more critical factor influencing AI Mocap performance than movement complexity. The results of this study provide practical guidelines for animators in selecting optimal filming angles to obtain high-quality motion data from AI Mocap systems for further application.
References
Baumgartner, T. (2024). Deriving sport scientific insights from broadcast footage: Advanced computer vision for human motion analysis [Doctoral dissertation, German Sport University]. https://bit.ly/4tgczbN
Güler, R., Neverova, N., & Kokkinos, I. (2020). DensePose: Dense human pose estimation in the wild. International Journal of Computer Vision, 128(2), 173–192. https://doi.org/10.1007/s11263-019-01202-8
Habibie, I., Holden, D., Schwarz, J., Jahrmann, M., & Theobalt, C. (2021). Learning realistic human motion from videos using adversarial networks. ACM Transactions on Graphics, 40(4), 1–13. https://doi.org/10.1145/3450626.3459794
Ishac, K., & Eager, D. (2021). Evaluating martial arts punching kinematics using a vision and inertial sensing system. Sensors, 21(6), 1948. https://doi.org/10.3390/s21061948
Ling, H., Zhang, C., Xu, H., & Liu, C. K. (2023). Physics-aware motion retargeting with whole-body coordination. IEEE Transactions on Visualization and Computer Graphics, 29(4), 2456–2470. https://doi.org/10.1109/TVCG.2022.3189021
Li, Q., Sun, T., & Zhang, M. (2025). Deep learning-driven animation: Enhancing real-time character motion synthesis. IEEE Access, 13, 181487-181506. https://doi.org/10.1109/ACCESS.2025.3623285
Luo, W., Chen, X., & Liu, Y. (2023). Motion transformer: Seamless transition of human poses in animation. Computer Animation and Virtual Worlds, 34(3–4), e2053. https://doi.org/10.1002/cav.2053
Ma, H., & Chen, G. (2022). Research trends and frontiers in exercise for movement disorders: A bibliometric analysis of global research from 2010 to 2021. Frontiers in Public Health, 10, 9491729. https://doi.org/10.3389/fpubh.2022.9491729
Ma, J., & Wang, Y. (2020). Comparison of a Marker-based optical motion capture system and a markerless vision-based system for evaluating human movement. Scientific Reports, 10(1), 17351. https://doi.org/10.1038/s41598-020-74567-9
Mathis, A., Schneider, S., Lauer, J., & Mathis, M. W. (2020). A Primer on Motion Capture with Deep Learning: Principles, Pitfalls, and Perspectives. Neuron, 108(1), 44–65. https://doi.org/10.1016/j.neuron.2020.09.017
Menache, A. (2011). Understanding motion capture for computer animation. Elsevier Science
Move Ai Ltd. (2025). Industry leading markerless motion capture. https://move.ai/
Noiumkar, S., & Tirakoat, S. (2013). Use of optical motion capture in sports science: A case study of golf swing. In N. Mai, M. Zamani, N. A. Ahmad@Salleh & W. C. Onn (Eds.), Proceedings - 2013 International Conference on Informatics and Creative Multimedia, ICICM 2013 (pp. 310–313). IEEE.
Pavllo, D., Feichtenhofer, C., Grangier, D., & Auli, M. (2019). 3D human pose estimation in video with temporal convolutions and semi-supervised training. In L. Davis, P. Torr & S. Zhu (Eds.), Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (pp. 7753–7762). IEEE.
Phalapree, W., & Tirakoat, S. (2024). Acceptance of artificial intelligence motion capture technology in animation production [In Thai]. Journal of Applied Information Technology, 10(2), 89–101. https://ph02.tci-thaijo.org/index.php/project-journal/article/view/253829
Potisarn, N. (2023). Applying electromagnetic motion capture equipment to create industrial 3D animations [In Thai]. Journal of Engineering and Industrial Technology, Kalasin University, 1(1), 1–9. https://doi.org/10.14456/jeit.2023.27
Robertson, D. G. E., Caldwell, G. E., Hamill, J., Kamen, G., & Whittlesey, S. N. (2013). Research methods in biomechanics (2nd ed.). Human Kinetics.
Starke, S., Zhao, Y., Zinno, F., & Komura, T. 2021. Neural animation layering for synthesizing martial arts movements. ACM Transactions on Graphics, 40(4), 1–16. https://doi.org/10.1145/3450626.3459881
Wongbat, S., Uypatchawong, S., & Chanamarn, N. (2024). The development of knowledge transfer application of ancient boxing dance in Sakon Nakhon Province [In Thai]. EAU Heritage Journal Science and Technology, 18(2), 43–58. https://he01.tci-thaijo.org/index.php/EAUHJSci/article/view/267368
Wu, J., Liu, H., Zhang, Y., & Chen, T. (2021). Efficient human pose estimation for action recognition using monocular videos. IEEE Transactions on Multimedia, 23, 2190–2203. https://doi.org/10.1109/TMM.2021.3051163
Zhang, X., Wang, L., Chen, M., & Li, Z. (2023). Viewpoint-robust 3D human pose estimation via depth-guided graph learning. IEEE Transactions on Image Processing, 32, 2941–2954. https://doi.org/10.1109/TIP.2023.3248271
