本研究旨在針對自主訓練情境下壺鈴硬舉常見的錯誤動作辨識問題，提出一套基於單視角影片的深度學習動作品質分類架構。相較於過去方法或仰賴多視角系統與動作捕捉設備而建置成本較高，或僅使用單一視角影片而對視角變異泛化能力不足，本研究以單一視角影片所估計之二維人體關鍵點與角度為輸入，結合代理生物力學建模，以提升模型對動作機制的表徵能力。本研究共包含四個多標籤分類任務。其中，髖過早啟動與膝主導之辨識由兩階段模型 TS-HK 處理:第一階段以二維關鍵點與關節角度預測三維關鍵點、關節角度與關節接觸力等代理特徵，第二階段再融合原始二維特徵與三維代理特徵進行分類;圓背則另以由影像分割取得之背部曲率特徵建模的 XGB-R 進行辨識。結果顯示，整合圓背獨立建模、投影式二維資料增強、三維運動學與力學特徵之 TS-HK + XGB-R 模型表現最佳，其 Macro-F1 與 Macro-AUC 分別達 0.7811 與 0.9619。研究結果指出，偵測圓背動作宜採任務專屬的形態特徵獨立建模，並且包含生物力學資訊的代理特徵導入亦有助於提升髖過早啟動的 F1 分數與整體 Macro-F1 分數表現。

This study aims to address the problem of identifying common erroneous movement patterns in kettlebell deadlifts during self-training by proposing a deep learning framework for movement quality classification based on single-view video. Compared with previous approaches that either rely on multi-view systems and motion capture devices with high deployment cost, or use only single-view video but suffer from limited generalization to viewpoint variation, the proposed method takes two-dimensional human keypoints and joint angles estimated from single-view video as input and incorporates surrogate biomechanical modeling to enhance the representation of movement mechanisms. This study includes four multi-label classification tasks. The identification of hip-early-rise and knee-dominant patterns is handled by a two-stage model, TS-HK: in the first stage, two-dimensional keypoints and joint angles are used to predict surrogate features, including three-dimensional joint coordinates, joint angles, and joint contact forces; in the second stage, the original two-dimensional features are fused with the three-dimensional surrogate features for classification. Rounded-back is identified by a separate model, XGB-R, built on back-curvature features extracted from image segmentation. The results show that the TS-HK + XGB-R model, which integrates independent rounded-back modeling, projection-based two-dimensional data augmentation, and surrogate 3D kinematic and kinetic features, achieved the best performance, with Macro-F1 and Macro-AUC values of 0.7811 and 0.9619, respectively. The findings indicate that rounded-back detection benefits from independent modeling using task-specific morphological features. Moreover, incorporating surrogate features that contain biomechanical information improves not only the class-specific F1-score for hip early rise, but also the overall Macro-F1 performance.

Bengtsson, V., Berglund, L., and Aasa, U. (2018). Narrative review of injuries in powerlifting with special reference to their association to the squat, bench press and deadlift. BMJ open sport & exercise medicine, 4(1).
Burton II, W. S., Myers, C. A., and Rullkoetter, P. J. (2021). Machine learning for rapid estimation of lower extremity muscle and joint loading during activities of daily living. Journal of Biomechanics, 123:110439.
Cao, Z., Hidalgo, G., Simon, T., Wei, S.-E., and Sheikh, Y. (2019). Openpose: Realtime multi-person 2d pose estimation using part affinity fields. IEEE transactions on pattern analysis and machine intelligence, 43(1):172–186.
Chen, C.-H. and Ramanan, D. (2017). 3d human pose estimation= 2d pose estimation+ matching. In Proceedings of the IEEE conference on computer vision and pattern recog- nition, pages 7035–7043.
Chen, C.-H., Tyagi, A., Agrawal, A., Drover, D., Mv, R., Stojanov, S., and Rehg, J. M. (2019). Unsupervised 3d pose estimation with geometric self-supervision. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 5714–5724.
Chen, T. and Guestrin, C. (2016). Xgboost: A scalable tree boosting system. In Proceedings of the 22nd acm sigkdd international conference on knowledge discovery and data mining, pages 785–794.
Delp, S. L., Anderson, F. C., Arnold, A. S., Loan, P., Habib, A., John, C. T., Guendelman, E., and Thelen, D. G. (2007). Opensim: open-source software to create and analyze dynamic
simulations of movement. IEEE transactions on biomedical engineering, 54(11):1940– 1950.
Fisher, J., Steele, J., Wolf, M., Korakakis, P. A., Smith, D., and Giessing, J. (2022). The role of supervision in resistance training; an exploratory systematic review and meta-analysis: Supervision in resistance training. International Journal of Strength and Conditioning, 2(1).
Fu, H., Gao, J., and Liu, H. (2023). Human pose estimation and action recognition for fitness movements. Computers & Graphics, 116:418–426.
Gamra, M. B. and Akhloufi, M. A. (2021). A review of deep learning techniques for 2d and 3d human pose estimation. Image and Vision Computing, 114:104282.
Haberkamp, L. D., Garcia, M. C., and Bazett-Jones, D. M. (2022). Validity of an artifi- cial intelligence, human pose estimation model for measuring single-leg squat kinematics. Journal of Biomechanics, 144:111333.
Keogh, J. W. and Winwood, P. W. (2017). The epidemiology of injuries across the weight- training sports. Sports medicine, 47(3):479–501.
Kotte, H., Kravcík, M., and Duong-Trung, N. (2023). Real-time posture correction in gym exercises: A computer vision-based approach for performance analysis, error classification and feedback. In MILeS@ EC-TEL, pages 64–70.
Li, C. and Lee, G. H. (2019). Generating multiple hypotheses for 3d human pose estimation with mixture density network. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 9887–9895.
Liew, B. X., Rügamer, D., Mei, Q., Altai, Z., Zhu, X., Zhai, X., and Cortes, N. (2023). Smooth and accurate predictions of joint contact force time-series in gait using over parameterised deep neural networks. Frontiers in Bioengineering and Biotechnology, 11:1208711.
Mazzetti, S. A., Kraemer, W. J., Volek, J. S., Duncan, N. D., Ratamess, N. A., Gómez, A. L., Newton, R. U., Hakkinen, K., and Fleck, S. J. (2000). The influence of direct supervision of resistance training on strength performance. Medicine and science in sports and exercise, 32(6):1175–1184.
Nogueira, A. F. R., Oliveira, H. P., and Teixeira, L. F. (2025). Markerless multi-view 3d human pose estimation: A survey. Image and Vision Computing, 155:105437.
Poursaberi, A., Bahr, C., Pluk, A., Van Nuffel, A., and Berckmans, D. (2010). Real-time automatic lameness detection based on back posture extraction in dairy cattle: Shape anal- ysis of cow with image processing techniques. Computers and electronics in agriculture, 74(1):110–119.
Rahmadani, A., Dewantara, B. S. B., and Sari, D. M. (2022). Human pose estimation for fitness exercise movement correction. In 2022 International Electronics Symposium (IES), pages 484–490. IEEE.
Shahzad, H., Jabeen, H., Arslan, H. R. M., Ghouri, M. H., Ali, S., Gondal, M. U., Arif, M., and Bibi, S. (2021). Musculoskeletal injuries among weight lifters with or without supervision? a comparative cross sectional study. Age, 15:35.
TensorFlow Authors (2021). Tensorflow.js pose detection. https://github.com/ tensorflow/tfjs-models/blob/master/pose-detection/README.md.
Uhlrich, S. D., Falisse, A., Kidziński, Ł., Muccini, J., Ko, M., Chaudhari, A. S., Hicks, J. L., and Delp, S. L. (2023). Opencap: Human movement dynamics from smartphone videos. PLoS computational biology, 19(10):e1011462.
Ultralytics (2023). Ultralytics yolov8.
Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., Kaiser, Ł., and Polosukhin, I. (2017). Attention is all you need. Advances in neural information processing systems, 30.
Winwood, P. W., Hume, P. A., Cronin, J. B., and Keogh, J. W. (2014). Retrospective injury epidemiology of strongman athletes. The Journal of Strength & Conditioning Research, 28(1):28–42.
Yasin, H., Iqbal, U., Kruger, B., Weber, A., and Gall, J. (2016). A dual-source approach for 3d pose estimation from a single image. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 4948–4956.
Zhao, L., Peng, X., Tian, Y., Kapadia, M., and Metaxas, D. N. (2019). Semantic graph convolutional networks for 3d human pose regression. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 3425–3435.
Zhao, R., Wang, Y., and Martinez, A. M. (2017). A simple, fast and highly-accurate algorithm to recover 3d shape from 2d landmarks on a single image. IEEE transactions on pattern analysis and machine intelligence, 40(12):3059–3066.
Zhao, Z., Kiciroglu, S., Vinzant, H., Cheng, Y., Katircioglu, I., Salzmann, M., and Fua, P. (2022). 3d pose based feedback for physical exercises. In Proceedings of the Asian Conference on Computer Vision, pages 1316–1332.