在台灣關於職業安全以及職業傷害已有詳細規範及法規，雇主須依循職業安全衛生法所規定之工作場所以及作業內容去進行調整，在評估職業傷害時業界大多採用國際通用量表例如REBA、OWAS、EWAS、RULA等等，目前使用這些量表時主要是透過評估者觀察受測者工作時的姿勢來進行評分，此方法相當耗費人力以及時間，在角度計算上如果單純用眼睛主觀估計可能不夠精確，而用精密儀器設備進行量測可能造成受測者的不方便或不自然。
影像處理之動作分析技術是現今常用來記錄並處理人體動作的一套技術，它可以應用於各大領域例如：體育、醫療、工業、機器人領域以及娛樂等等。此技術的內容主要是利用攝影機或感應器擷取人體關節位置並進行運算，進而取得人體關節角度以及動作的數據。利用這些數據就可以分析人在做各種行為時的身體動作，找出更有效率的運動方式並提出一套動作策略提高工作效率。
因此本研究以建立出一套利用影像分析進行職業傷害評估系統為目的，利用GoPro HERO 6 Black攝影機拍攝作業內容並將彩色影像輸入OpenPose進行人體姿勢運算，利用運算結果計算人體關節角度進而配合REBA量表評分，最終將量表分數輸出告知使用者此作業內容之肌肉骨骼危害程度。
本研究之自動化量表評估系統之角度誤差驗證結果，整體絕對誤差為5.04度，整體線性誤差為-3.17度，達到規範之線性誤差正負5度以內（Norkin & White, 2016）。與專業人員評估之比較結果，頭部、軀幹、腿部、上臂、前臂與手腕平均分數相差0.33、0.75、0.03、0.03、0.08、0分，姿勢分數A、B、C與REBA總體平均分數相差1.10、0.03、0.75、0.75分。
自動化量表評估系統透過角度驗證與專業人員評估比較後證實了可信度與可行性，自動化量表系統能正確評估各關節角度避免肉眼評估誤差，也大幅縮減評估所需耗費的時間（62.5%，自動化量表評估系統全影格評估比較專業人員評估十分鐘影片非全影格評估）以及人力成本，並能長時間監控勞工作業姿勢，達到預防職業傷害的發生。

In Taiwan, there are detailed regulations on occupational safety and occupational injuries. Employers must follow the Occupational Safety and Health Law to design the workplace and work content. When assessing occupational injuries, most of the industry uses internationally accepted assessment scales such as REBA, OWAS, EWAS, and RULA. The current use of these assessment scales is mainly through observing the posture of the subject at work. This method is quite labor intensive and time-consuming. Using visual estimation may not be accurate enough when calculating joint angles, and using precision instruments may cause inconvenience or unnaturalness to the subject.
Motion analysis by image processing is a technique that is commonly used today to record and analyze human motion. It can be applied to various fields such as sports, medical treatments, industrial, robotics, and entertainment. The content of this technology is to use the camera or sensor to capture the position of the human joint and calculate the joint angle and motion . Using the data, you can analyze the body movements of people doing various behaviors, find out more efficient ways of exercising, and provide a set of action strategies to improve work efficiency.
This study aims to establish an automated system that evaluates occupational injury by using image processing. The GoPro HERO 6 Black camera is used to capture the content of the work, and the color image is input into OpenPose for human posture calculation. The calculation results are used to calculate the angle of the human joint and the score of REBA. Finally, the score of REBA is used to inform the user of the musculoskeletal risk of the work content.
The angular error verification results of the automatic evaluation system have an average absolute error of 5.04 degrees and an average algebraic error of -3.17 degrees. Compared with the evaluation of a professional, the average difference of the head, trunk, leg, upper arm, lower arm, and wrist are 0.33, 0.75, 0.03, 0.03, 0.08, 0 points respectively, and the average difference of posture score A, B, C and REBA score are 1.10, 0.03, 0.75, 0.75 points respectively.
The automatic assessment scale evaluation system verifies the reliability and acceptability by doing the angle verification and comparing with the evaluation of a professional. The system can correctly evaluate the joint angle to avoid visual evaluation error, significantly reduce the time and labor cost required for the evaluation. Furthermore, it can also monitor the labor’s work posture for a long time to prevent the occurrence of occupational injuries.

目錄
第一章 緒論..........8
1.1. 研究背景與動機..........8
1.2. 研究目的與範圍..........14
1.3. 研究架構與流程..........15
第二章 文獻回顧..........16
2.1. 不適當姿勢造成的肌肉骨骼影響..........16
2.2. 量表評估法..........20
2.2.1. 快速全身肢體評估法（REBA）..........20
2.2.2. 勞工姿勢評估系統（OWAS）..........28
2.2.3. 快速上肢評估法（RULA）..........32
2.2.4. 歐洲人因評估工具檢核表（EAWS）..........34
2.2.5. 小結..........38
2.3. 動作擷取系統..........39
2.3.1. 非影像式動作擷取系統..........39
2.3.2. 影像式動作分析系統..........40
2.4. 影像處理用於動作分析之計算技術..........41
2.4.1. MPII Human Pose Models..........48
2.4.2. OpenPose..........53
2.5. 小結..........56
第三章 研究方法..........57
3.1. 問題定義與描述..........57
3.2. 資料獲取與系統架構..........57
3.2.1. 資料獲取..........58
3.3. 研究設置..........60
3.4. 統計分析..........67
第四章 實驗結果..........68
4.1. 角度驗證成果..........68
4.1.1. VICON系統標準貼法之驗證作業..........68
4.1.2. OpenPose系統貼法之驗證作業..........75
4.2. 模擬工作姿勢與REBA量表進行比較..........83
第五章 討論..........90
5.1. 角度驗證成果..........90
5.2. 自動化量表評估系統於REBA量表之成效..........91
第六章 結論與未來研究方向..........96
參考文獻..........97

1.勞動部勞動及職業安全衛生研究所，「工作環境安全衛生狀況認知調查-2013年」。勞安所研究報告(IOSH102-M306)，98-102，2013。
2.勞動部職業安全衛生署（2018），職業病鑑定。
取自：https://www.osha.gov.tw/1106/1176/1185/1186/2623/
3.行政院勞委會勞工安全衛生研究所（2008），勞工衛生與職業病預防概論。取自：https://www.ilosh.gov.tw/menu/1223/1235/1237/%E5%8B%9E%E5%B7%A5%E5%AE%89%E5%85%A8%E8%88%87%E8%81%B7%E6%A5%AD%E5%82%B7%E5%AE%B3%E9%A0%90%E9%98%B2%E6%A6%82%E8%AB%96/
4.勞動部職業安全衛生署（2015），「人因性危害預防計畫指引」。
取自：https://www.osha.gov.tw/1106/1251/10159/10173/10309/
5.職業安全衛生法 - 勞動部勞動法令查詢系統-所有條文（2013），職業安全衛生法。取自：http://laws.mol.gov.tw/flaw/FLAWDAT0201.aspx?lsid=FL015013
6.以 SOWT 分析光學式動作擷取系統發展趨勢. 2010 年第三屆運動科學暨休閒遊憩管理學術研討會論文集, 2011.
7.詳解慣性動作捕捉技術的應用領域（2017）。
取自：https://hk.saowen.com/a/6e0c5a12156d7f0c386d1be0d2d5dbe43303492faf409c3254b753cd6810c659

8.勞動部勞動及職業安全衛生研究所（2015），「肌肉骨骼健康狀態之職場相關因子 關聯性探討」。
取自：https://labor-elearning.mol.gov.tw/base/10001/door/%E5%A0%B1%E5%91%8A%E5%8D%80/795_ILOSH103-H504%E8%82%8C%E8%82%89%E9%AA%A8%E9%AA%BC%E5%81%A5%E5%BA%B7%E7%8B%80%E6%85%8B%E4%B9%8B%E8%81%B7%E5%A0%B4%E7%9B%B8%E9%97%9C%E5%9B%A0%E5%AD%90%E9%97%9C%E8%81%AF%E6%80%A7%E6%8E%A2%E8%A8%8E.pdf
9.勞動部勞動及職業安全衛生研究所（2014），「人因工程肌肉骨骼傷病預防指引」。
取自：http://laws.ilosh.gov.tw/ioshcustom/Web/TechPublications/Detail?id=135
10.武文孝（2016）。越南家事清潔人員累積性肌肉骨骼傷害之研究（碩士論文）。取自：http://ir.lib.kuas.edu.tw/handle/987654321/16061
11.林慕宗（2017）。肌肉骨骼危害防制實物探討-以面板製造為例（碩士論文）。取自：https://ir.lib.ntut.edu.tw/wSite/ct?mp=ntut&xItem=68996&ctNode=447

外文部分：
1.Baek, S., & Kim, M. (2016). Real-Time Dynamic Motion Capture Using Multiple Kinects. In Advances in Computer Science and Ubiquitous Computing (pp. 29-35): Springer.
2.Cao, Z., Simon, T., Wei, S.-E., & Sheikh, Y. (2017). Realtime multi-person 2d pose estimation using part affinity fields. Paper presented at the Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.
3.Chowdhury, S. S., Boricha, J., & Yardi, S. (2012). Identification of awkward postures that cause discomfort to Liquid Petroleum Gas workers in Mumbai, India. Indian journal of occupational and environmental medicine, 16(1), 3.
4.Coenen, P., Kingma, I., Boot, C. R., Bongers, P. M., & van Dieën, J. H. (2013). Inter-rater reliability of a video-analysis method measuring low-back load in a field situation. Applied ergonomics, 44(5), 828-834.
5.Coenen, P., Kingma, I., Boot, C. R., Faber, G. S., Xu, X., Bongers, P. M., & Van Dieen, J. H. (2011). Estimation of low back moments from video analysis: A validation study. Journal of biomechanics, 44(13), 2369-2375.
6.Genaidy, A., Simmons, R. J., Guo, L., & Hidalgo, J. (1993). Can visual perception be used to estimate body part angles? Ergonomics, 36(4), 323-329.
7.Hignett, S., & McAtamney, L. (2000). Rapid entire body assessment (REBA). Applied ergonomics, 31(2), 201-205.
8.Insafutdinov, E., Pishchulin, L., Andres, B., Andriluka, M., & Schiele, B. (2016). Deepercut: A deeper, stronger, and faster multi-person pose estimation model. Paper presented at the European Conference on Computer Vision.
9.Iqbal, U., & Gall, J. (2016). Multi-person pose estimation with local joint-to-person associations. Paper presented at the European Conference on Computer Vision.
10.Kivi, P., & Mattila, M. (1991). Analysis and improvement of work postures in the building industry: application of the computerised OWAS method. Applied ergonomics, 22(1), 43-48.
11.Krüger, J., & Nguyen, T. D. (2015). Automated vision-based live ergonomics analysis in assembly operations. CIRP Annals, 64(1), 9-12.
12.Malaisé, A., Maurice, P., Colas, F., & Ivaldi, S. (2019). Activity Recognition for Ergonomics Assessment of Industrial Tasks with Automatic Feature Selection. IEEE Robotics and Automation Letters, 4(2), 1132-1139.
13.Mattila, M., Karwowski, W., & Vilkki, M. (1993). Analysis of working postures in hammering tasks on building construction sites using the computerized OWAS method. Applied ergonomics, 24(6), 405-412.
14.Moussavi-Najarkola, S. A., & Mirzaei, R. (2012). Assessment of musculoskeletal loads of electric factory workers by rapid entire body assessment. Health Scope, 1(2), 71-79.
15.Norkin, C. C., & White, D. J. (2016). Measurement of joint motion: a guide to goniometry: FA Davis.
16.Pishchulin, L., Insafutdinov, E., Tang, S., Andres, B., Andriluka, M., Gehler, P. V., & Schiele, B. (2016). Deepcut: Joint subset partition and labeling for multi person pose estimation. Paper presented at the Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.
17.Roquelaure, Y., Ha, C., Rouillon, C., Fouquet, N., Leclerc, A., Descatha, A., ... & Members of Occupational Health Services of the Pays de la Loire Region. (2009). Risk factors for upper‐extremity musculoskeletal disorders in the working population. Arthritis Care & Research, 61(10), 1425-1434.
18.Sementille, A., Escaramuzi Lourenço, L., Brega, J. R., & Rodello, I. (2004). A motion capture system using passive markers.
19.Sen, A., & Richardson, S. (2007). A study of computer-related upper limb discomfort and computer vision syndrome. Journal of human ergology, 36(2), 45-50.
20.Syahril, F., & Sonjaya, E. (2015). Validity, sensitivity, and relia⁃ bility testing by ergonomic evaluation methods for geothermal task. Paper presented at the Proceedings World Geothermal Congress.
21.Vander Linden, D. W., Carlson, S. J., & Hubbard, R. L. (1992). Reproducibility and accuracy of angle measurements obtained under static conditions with the Motion Analysis™ video system. Physical Therapy, 72(4), 300-305.
22.Walther, M., & Muñoz, B. T. (2012). Integration of time as a factor in ergonomic simulation. Work, 41(Supplement 1), 4372-4375.
23.Wei, S.-E., Ramakrishna, V., Kanade, T., & Sheikh, Y. (2016). Convolutional pose machines. Paper presented at the Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.
24.Zhang, Z. (2000). A flexible new technique for camera calibration. IEEE Transactions on pattern analysis and machine intelligence, 22.