研究生: |
張育維 Yu-Wei Chang |
---|---|
論文名稱: |
使用深度學習技術檢測運動傷害 Using Deep Learning Techniques to Detect Sports Injuries |
指導教授: |
洪西進
Shi-Jinn Horng |
口試委員: |
洪西進
Shi-Jinn Horng 謝仁偉 Jen-Wei Hsieh 楊竹星 Chu-Sing Yang 李正吉 Cheng-Chi Lee 林祝興 Chu-Hsing Lin |
學位類別: |
碩士 Master |
系所名稱: |
電資學院 - 資訊工程系 Department of Computer Science and Information Engineering |
論文出版年: | 2022 |
畢業學年度: | 110 |
語文別: | 中文 |
論文頁數: | 53 |
中文關鍵詞: | 運動損傷 、姿態預測 |
外文關鍵詞: | Sport Injuries, Pose Estimation |
相關次數: | 點閱:253 下載:0 |
分享至: |
查詢本校圖書館目錄 查詢臺灣博碩士論文知識加值系統 勘誤回報 |
本論文依據現有的研究結果作為根據,找出三種可能造成運動傷害的指標,並依據這三種指標分別建立三種辨識運動異常模型,先透過OpenPose[1](Z. Cao et al., 2019)提取人體關節點座標和感興趣的特徵,再經過異常值處理後輸入至本論文的辨識運動異常模型,最後判斷運動姿勢是否可能造成運動損傷?
三種運動傷害指標包含 :
(1) 跑步時膝蓋是否有外翻和內翻?
(2) 在跑步過程中,大腿蹬地往前跳時是否髖關節伸展不足?
(3) 深蹲時膝蓋是否有外翻的現象?
這三種運動表現特徵皆有可能造成膝關節傷害,數據收集會依據量測指標的不同收集不同角度的影片,指標(1)選擇相機由前方或後方拍攝跑者的影片;指標(2)選擇相機從側方拍攝跑者的影片;指標(3)選擇從前方拍攝使用者做深蹲訓練的影片。從youtube收集和自己拍攝這三種視角的影片作為本研究的資料集。本研究根據這3個運動傷害指標分別建立的運動傷害辨識模型分別達到了94.1%、93.7%和95.4%的準確度。
運動傷害模型訓練完成後和OpenPose[1](Cao et al., 2019)做整合,實作了運動傷害即時檢測系統。同時設計一個簡單易操作的系統介面,可以透過攝影機或錄製好的影片作為輸入,提供使用者隨時檢測運動姿勢是否正確,判斷是否可能會造成運動傷害?
Based on the existing research results, we find three indicators that may cause sports injuries, and we establish three models to identify abnormal motion based on these three indicators. First, we use OpenPose[1](Cao et al., 2019) as the human joint extractor in this study to obtain the joint point coordinates and the features of interest, and then input them into the proposed abnormal motion recognition model to determine whether the motion posture may cause sports injury after outlier processing.
The three sports indicators include:
(1) the knees are genu valgum or baker-kneed while running
(2) whether the hip joint is insufficiently stretched when the thighs jump forward in the late running posture
(3) the knees are genu valgum while squatting
According to the existing researches, these three kinds of motion representations may cause knee injury. The data collection method will collect videos from different angles according to different measurement indicators. Indicator (1) chooses videos recording runners running from front or back perspective; indicator (2) chooses videos recording runners running from side perspective; indicator (3) chooses videos recording user doing squat training from front side perspective. All videos were collected from Youtube or recorded by myself based on three perspectives as the data set for this research. Three models established in this study based on these three sports injury indicators achieved 94.1%, 93.7% and 95.4% accuracy respectively.
After the training of the sports injury model is completed, it is integrated with OpenPose[1](Cao et al., 2019), and we implement real-time sports injury detection system, a simple and easy-to-operate system, which can use camera stream or video as input to input into system, so that users can check whether they may cause sports injury.
[1] Z. Cao, G. Hidalgo, T. Simon, S.-E. Wei, and Y. Sheikh, “Openpose: realtime multi-person 2D pose estimation using part affinity fields,” arXiv, arXiv:1812.08008, May 2019. doi: 10.48550/arXiv.1812.08008.
[2] D. R. Patel and A. Villalobos, “Evaluation and management of knee pain in young athletes: overuse injuries of the knee,” Transl Pediatr, vol. 6, no. 3, pp. 190–198, Jul. 2017, doi: 10.21037/tp.2017.04.05.
[3] J.-F. Esculier, T. Silvini, L. J. Bouyer, and J.-S. Roy, “Video-based assessment of foot strike pattern and step rate is valid and reliable in runners with patellofemoral pain,” Phys Ther Sport, vol. 29, pp. 108–112, Jan. 2018, doi: 10.1016/j.ptsp.2016.11.003.
[4] B. Dingenen, C. Barton, T. Janssen, A. Benoit, and P. Malliaras, “Test-retest reliability of two-dimensional video analysis during running,” Phys Ther Sport, vol. 33, pp. 40–47, Sep. 2018, doi: 10.1016/j.ptsp.2018.06.009.
[5] B. Dingenen et al., “Are two-dimensional measured frontal plane angles related to three-dimensional measured kinematic profiles during running?,” Phys Ther Sport, vol. 29, pp. 84–92, Jan. 2018, doi: 10.1016/j.ptsp.2017.02.001.
[6] M. L. Gorman McNerney and E. A. Arendt, “Anterior knee pain in the active and athletic adolescent,” Curr Sports Med Rep, vol. 12, no. 6, pp. 404–410, Dec. 2013, doi: 10.1249/JSR.0000000000000013.
[7] R. B. Souza, “An evidence-based videotaped running biomechanics analysis,” Phys Med Rehabil Clin N Am, vol. 27, no. 1, pp. 217–236, Feb. 2016, doi: 10.1016/j.pmr.2015.08.006.
[8] T. E. Hewett et al., “Biomechanical measures of neuromuscular control and valgus loading of the knee predict anterior cruciate ligament injury risk in female athletes: a prospective study,” Am J Sports Med, vol. 33, no. 4, pp. 492–501, Apr. 2005, doi: 10.1177/0363546504269591.
[9] T. E. Hewett, A. L. Stroupe, T. A. Nance, and F. R. Noyes, “Plyometric training in female athletes. Decreased impact forces and increased hamstring torques,” Am J Sports Med, vol. 24, no. 6, pp. 765–773, Dec. 1996, doi: 10.1177/036354659602400611.
[10] A. Heijink et al., “Biomechanical considerations in the pathogenesis of osteoarthritis of the knee,” Knee Surg Sports Traumatol Arthrosc, vol. 20, no. 3, pp. 423–435, 2012, doi: 10.1007/s00167-011-1818-0.
[11] D. G. Lloyd and T. S. Buchanan, “Strategies of muscular support of varus and valgus isometric loads at the human knee,” J Biomech, vol. 34, no. 10, pp. 1257–1267, Oct. 2001, doi: 10.1016/s0021-9290(01)00095-1.
[12] H.-L. Teng and C. M. Powers, “Hip-extensor strength, trunk posture, and use of the knee-extensor muscles during running,” J Athl Train, vol. 51, no. 7, pp. 519–524, Jul. 2016, doi: 10.4085/1062-6050-51.8.05.
[13] J. L. Elman, “Finding structure in time,” Cognitive Science, vol. 14, no. 2, pp. 179–211, 1990, doi: 10.1207/s15516709cog1402_1.
[14] S. Hochreiter and J. Schmidhuber, “Long short-term memory,” Neural Computation, vol. 9, no. 8, pp. 1735–1780, Nov. 1997, doi: 10.1162/neco.1997.9.8.1735.
[15] V. Bazarevsky, I. Grishchenko, K. Raveendran, T. Zhu, F. Zhang, and M. Grundmann, “Blazepose: on-device real-time body pose tracking,” arXiv, arXiv:2006.10204, Jun. 2020. doi: 10.48550/arXiv.2006.10204.
[16] A. Newell, K. Yang, and J. Deng, “Stacked hourglass networks for human pose estimation,” arXiv, arXiv:1603.06937, Jul. 2016. doi: 10.48550/arXiv.1603.06937.
[17] K. Simonyan and A. Zisserman, “Very deep convolutional networks for large-scale image recognition,” arXiv, arXiv:1409.1556, Apr. 2015. doi: 10.48550/arXiv.1409.1556.
[18] F. Karim, S. Majumdar, H. Darabi, and S. Chen, “LSTM fully convolutional networks for time series classification,” IEEE Access, vol. 6, pp. 1662–1669, 2018, doi: 10.1109/ACCESS.2017.2779939.