LEAKAGE-FREE CAUSAL CONTEXT REPRESENTATION FOR POSE-BASED EXERCISE PHASE RECOGNITION
https://doi.org/10.53360/2788-7995-2026-3(23)-24
Abstract
Pose-based exercise recognition provides a computationally efficient approach for analyzing human movement; however, classification based solely on the current pose may fail to capture shortterm motion context. This study investigates whether leakage-free causal context can improve exercise-phase recognition without requiring a complex temporal neural architecture. A time-series exercise dataset containing 1,372 ordered pose observations from five exercises and ten exercise-phase classes was analyzed using seven pose representations derived from body landmarks, joint angles, and spatial distances. To reduce the risk of optimistic evaluation caused by neighboring observations appearing across training and testing partitions, a contiguous block-aware validation protocol was employed. A leakage-free causal mean context representation was constructed using only the current and preceding observations, without future samples or ground-truth labels. History-length ablation showed that a five-observation causal history achieved the best performance. The proposed representation increased mean Macro-F1 from 0.8358 for current-pose classification to 0.8973, corresponding to a 7.36% relative improvement, and improved performance in all five evaluation blocks. A causal Temporal Convolutional Network achieved a Macro-F1 of 0.8699, indicating that the simpler causal aggregation was more effective under the evaluated dataset conditions. Paired analysis yielded a mean Macro-F1 improvement of 0.0615 with a 95% confidence interval of [0.0308, 0.0921] and a large observed effect size (Cohen’s 𝑑𝑧 = 2.49); however, exact two-sided tests did not reach the conventional significance threshold (𝑝 = 0.0625) because only five paired blocks were available. These findings suggest that short causal pose context can provide useful information for exercise-phase recognition while remaining compatible with real-time processing.
About the Authors
G. N. KazbekovaKazakhstan
Gulnur Nagimetovna Kazbekova – Candidate of Technical Sciences, Associate Professor
29 Bekzat Sattarkhanov Avenue. Turkestan, Kazakhstan
Y. U. Serdaliyev
Kazakhstan
Yerlan Serdaliyev – PhD student in the educational program «Information systems»
29 Bekzat Sattarkhanov Avenue. Turkestan, Kazakhstan
References
1. Feng M. Skeleton Graph-Neural-Network-Based Human Action Recognition: A Survey / M. Feng, J. Meunier // Sensors. – 2022. – Vol. 22, № 6. – Art. 2091. https://doi.org/10.3390/s22062091.
2. Yue R. Action recognition based on RGB and skeleton data sets: A survey / R. Yue, Z.-Q. Tian, S. Du // Neurocomputing. – 2022. – Vol. 512. – P. 287-306. https://doi.org/10.1016/j.neucom.2022.09.071.
3. A Survey on 3D Skeleton-Based Action Recognition Using Learning Method / B. Ren et al // Cyborg and Bionic Systems. – 2024. – Vol. 5. – Art. 0100. https://doi.org/10.34133/cbsystems.0100.
4. Huang Q. A Review of Skeleton-Based Human Action Recognition / Q. Huang, J. Cui, C. Li // Journal of Computer-Aided Design & Computer Graphics. – 2024. – Vol. 36, № 2. – P. 173-194. https://doi.org/10.3724/SP.J.1089.2024.2023-00358.
5. Yan S. Spatial Temporal Graph Convolutional Networks for Skeleton-Based Action Recognition / S. Yan, Y. Xiong, D. Lin // Proceedings of the AAAI Conference on Artificial Intelligence. – 2018. – Vol. 32, № 1. https://doi.org/10.1609/aaai.v32i1.12328.
6. Channel-Wise Topology Refinement Graph Convolution for Skeleton-Based Action Recognition / Y. Chen et al // Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV). – 2021. – P. 13359-13368.
7. InfoGCN: Representation Learning for Human Skeleton-Based Action Recognition / H.-g. Chi еt al // Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). – 2022. – P. 20186-20196. https://doi.org/10.1109/CVPR52688.2022.01955.
8. Constructing Stronger and Faster Baselines for Skeleton-Based Action Recognition / Y.-F. Song et al // IEEE Transactions on Pattern Analysis and Machine Intelligence. – 2023. – Vol. 45, № 2. – P. 1474-1488. https://doi.org/10.1109/TPAMI.2022.3157033.
9. Nan M. Spatio-temporal neural network with handcrafted features for skeleton-based action recognition / M. Nan, M. Trăscău, A.-M. Florea // Neural Computing and Applications. – 2024. – Vol. 36. – P. 9221-9243. https://doi.org/10.1007/s00521-024-09559-4.
10. Temporal Convolutional Networks for Action Segmentation and Detection / C. Lea et al // Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR). – 2017. – P. 156-165.
11. Bai S. An Empirical Evaluation of Generic Convolutional and Recurrent Networks for Sequence Modeling / S. Bai, J. Z. Kolter, V. Koltun. – 2018. – arXiv:1803.01271.
12. Plizzari C. Skeleton-based action recognition via spatial and temporal transformer networks / C. Plizzari, M. Cannici, M. Matteucci // Computer Vision and Image Understanding. – 2021. – Vol. 208-209. – Art. 103219. https://doi.org/10.1016/j.cviu.2021.103219.
13. MotionBERT: A Unified Perspective on Learning Human Motion Representations / W. Zhu et al // Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV). – 2023. – P. 15085-15099.
14. Varga D. Critical Analysis of Data Leakage in WiFi CSI-Based Human Action Recognition Using CNNs / D. Varga // Sensors. – 2024. – Vol. 24, № 10. – Art. 3159. https://doi.org/10.3390/s24103159.
15. Recognition and Scoring Physical Exercises via Temporal and Relative Analysis of Skeleton Nodes Extracted from the Kinect Sensor / R. Esmaeeli et al // Sensors. – 2024. – Vol. 24, № 20. – Art. 6713. https://doi.org/10.3390/s24206713.
16. Efficient Skeleton-Based Action Recognition for Real-Time Embedded Systems / N. Noor et al // Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW). – 2024. – P. 5889-5897.
17. Varga D. Mitigating Data Leakage in a WiFi CSI Benchmark for Human Action Recognition / D. Varga // Sensors. – 2024. – Vol. 24, № 24. – Art. 8201. https://doi.org/10.3390/s24248201.
18. Liu J. Multi-Modality Co-Learning for Efficient Skeleton-based Action Recognition / J. Liu, C. Chen, M. Liu // Proceedings of the 32nd ACM International Conference on Multimedia. – 2024. – P. 4909-4918. https://doi.org/10.1145/3664647.3681015.
19. BlockGCN: Redefine Topology Awareness for Skeleton-Based Action Recognition / Y. Zhou et al // Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). – 2024. – P. 2049-2058. https://doi.org/10.1109/CVPR52733.2024.00200.
20. MGSAN: Multimodal Graph Self-Attention Network for Skeleton-Based Action Recognition / J. Wang et al // Multimedia Systems. – 2024. – Vol. 30. – Art. 369. https://doi.org/10.1007/s00530-024-01566-8.
21. FineRehab: A Multi-modality and Multi-task Dataset for Rehabilitation Analysis / J. Li et al // Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW). – 2024. – P. 3184-3193. https://doi.org/10.1109/CVPRW63382.2024.00324.
22. Towards Automating Personal Exercise Assessment and Guidance with Affordable Mobile Technology / M. Sideridou et al // Sensors. – 2024. – Vol. 24, № 7. – Art. 2037. https://doi.org/10.3390/s24072037.
23. Kwon H.-J. A novel machine learning-based shoulder exercise assistant program / H.-J. Kwon, S.-H. Choi // Discover Artificial Intelligence. – 2025. – Vol. 5. – Art. 37. https://doi.org/10.1007/s44163-025-00259-z.
24. M3GYM: A Large-Scale Multimodal Multi-view Multi-person Pose Dataset for Fitness Activity Understanding in Real-world Settings / Q. Xu et al // Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). – 2025. – P. 12289-12300.
25. A Knee Rehabilitation Exercises Dataset for Postural Assessment using Wearable Devices / P. Kasnesis et al // Scientific Data. – 2025. – Vol. 12. – Art. 610. https://doi.org/10.1038/s41597-025-04963-4.
Supplementary files
|
1. Неозаглавлен | |
| Subject | ||
| Type | Other | |
Download
(B)
|
Indexing metadata ▾ | |
Review
For citations:
Kazbekova G.N., Serdaliyev Y.U. LEAKAGE-FREE CAUSAL CONTEXT REPRESENTATION FOR POSE-BASED EXERCISE PHASE RECOGNITION. Bulletin of Shakarim University. Technical Sciences. 2026;(3(23)):236-248. https://doi.org/10.53360/2788-7995-2026-3(23)-24
JATS XML















