重量訓練是一種有效的健身方法，其能夠增強肌肉力量、提高新陳代謝和改善體態。而在每次訓練中記錄自己的運動過程與數據，能夠幫助個人規劃適當的訓練內容，提升運動效果，建立和維持健康的運動習慣。
在真實健身房中，開放式環境使得主要器材周圍會有許多其他器材和非訓練者，用於自動追蹤的攝影機，擺放位置受到諸多限制，無法放置於走道或離器材太近的地方，並且移動的非訓練者會對拍攝的訓練過程產生遮擋，影響訓練追蹤的效果。
本研究提出基於攝影機的單人機械式器材訓練追蹤方法，使用攝影機拍攝訓練者與訓練器材，透過人體姿態估計與物件偵測，獲得人體關鍵點與訓練器材的資訊，接著藉由獲得的資訊篩選出受到遮擋影響的關鍵點資訊進行過濾，再以KNN(k-Nearest Neighbor)插值法補償過濾掉的關鍵點，預測訓練者在遮擋時間的動作軌跡，根據補償過後的關鍵點資訊，推測出正確的動作次數。
本研究設計兩種實驗，分別檢驗單攝影機及多攝影機下補償方法的成效，實驗影片由三個視角同時拍攝，收集6個訓練者分別進行肩推、胸推、腿推的多部訓練影像，共180部影片。實驗結果顯示，在單一攝影機條件下，補償後的次數估計準確度較補償前提升5.6%，在多攝影機條件下，補償後的平均準確率可達98.9%，重量追蹤在不同視角下，平均準確率可達94.6%，綜合以上實驗結果，說明本研究提出的補償方法可以減少環境對於自動追蹤的干擾，提升追蹤準確度。

Weight training is an effective fitness method that can build muscle strength, increase metabolism, and improve posture. Recording one's own exercise process and data during each training can help individuals plan appropriate training content, improve exercise effects, and establish and maintain healthy exercise habits.
In actual gyms, the open environment results in many equipment and non-trainers surrounding the equipment. The position of the automatic tracking camera is limited, and moving non-trainers will block the shooting screen, affecting the training tracking results.
This research uses computer vision technology and uses cameras to capture trainers and training equipment. First, through human posture estimation and image recognition, information on key points of the human body and training equipment is obtained. Then, it is combined with motion state detection to screen out the key points affected by interference. Information. Then the KNN interpolation method is used to compensate the filtered key points, predict the actual movement trajectory of the trainer, and infer the correct number of movements based on the compensated key point information.
This study designed two experiments to test the effectiveness of the compensation method under single-view and multi-view respectively. The experimental videos were shot from three views at the same time, collecting multiple training images of 6 trainers performing shoulder press, chest press, and leg press respectively. 180 videos in total. Experimental results show that under the condition of a single camera, the accuracy of time estimation after compensation is 5.6% higher than before compensation. Under the condition of multiple cameras, the average accuracy after compensation can reach 98.9%. The weight tracking is averagely accurate under different viewing angles. The accuracy rate can reach 94.6%. Based on the above experimental results, it shows that the compensation method proposed in this study can reduce the interference of the environment on automatic tracking and improve the tracking accuracy.

[1] Freund, Y., & Schapire, R. E. (1997). A decision-theoretic generalization of on-line learning and an application to boosting. Journal of Computer and System Sciences, 55(1), 119–139. https://doi.org/10.1006/jcss.1997.1504
[2] Teng, G., Zhang, H., & Li, Y. (2020). MobiFit: Contactless fitness assistant for freehand exercises using just one cellular signal receiver. In Proceedings of the 16th International Conference on Mobility, Sensing and Networking (MSN). IEEE. https://doi.org/10.1109/MSN50589.2020.00057
[3] Park, H.-J., Baek, J.-W., & Kim, J.-H. (2022). Imagery-based parametric classification of correct and incorrect motion for push-up counter using OpenPose. In Proceedings of the IEEE 16th International Conference on Automation Science and Engineering (CASE). IEEE. https://doi.org/10.1109/CASE48305.2020.9216833
[4] Khurana, R., Zhang, Y., & Hsu, W. (2018). GymCam: Detecting, recognizing, and tracking simultaneous exercises in unconstrained scenes. Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies, 2(4), 1–17. https://doi.org/10.1145/3287063
[5] Bian, S., Wang, Y., & Zhang, L. (2022). Exploring automatic gym workouts recognition locally on wearable resource-constrained devices. In Proceedings of the IEEE 13th International Green and Sustainable Computing Conference (IGSC). IEEE. https://doi.org/ 10.1109/IGSC55832.2022.9969370
[6] Ultralytics. (2023). YOLOv8. Ultralytics Documentation. https://docs.ultralytics.com/
[7] Dedhia, U., Kumar, A., & Sharma, P. (2023). Pose estimation and virtual gym assistant using MediaPipe and machine learning. In Proceedings of the International Conference on Network, Multimedia and Information Technology (NMITCON). IEEE. https://doi.org/ 10.1109/NMITCON58196.2023.10275938
[8] Bazarevsky, V., Kim, K., & Zolotov, A. (2020). BlazePose: On-device real-time body pose tracking. In Proceedings of the CVPR Workshop on Computer Vision for Augmented and Virtual Reality (pp. 1–4). https://doi.org/10.48550/arXiv.2006.10204
[9] Li, X., Wang, Y., & Zhang, L. (2022). Fitness action counting based on MediaPipe. In Proceedings of the 15th International Congress on Image and Signal Processing, BioMedical Engineering and Informatics (CISP-BMEI). IEEE. https://doi.org/ 10.1109/CISP-BMEI56279.2022.9980337
[10]Cao, Z., Hidalgo, G., Simon, T., Wei, S.-E., & Sheikh, Y. (2018). OpenPose: Realtime multi-person 2D pose estimation using part affinity fields. arXiv preprint arXiv:1812.08008. https://arxiv.org/abs/1812.08008
[11]Zhang, Z., Wang, N., & Cui, L. (2018). Fine-Fit: A fine-grained gym exercises recognition system. In Proceedings of the 24th Asia-Pacific Conference on Communications (APCC). IEEE. https://doi.org/ 10.1109/APCC.2018.8633473
[12]林厚廷 (2024). 基於攝影機的自由重量訓練追蹤(碩士論文，國立臺灣師範大學)
[13]Lever Chest Press. (n.d.). Lift Manual. https://liftmanual.com/lever-chest-press/
[14]Lever Seated Shoulder Press. (n.d.). Lift Manual. https://liftmanual.com/lever-seated-shoulder-press/
[15]Sled 45 Leg Press. (n.d.). Lift Manual. https://liftmanual.com/sled-45-leg-press/