臺灣博碩士論文加值系統

本研究目的是探討AI影像辨識技術Keyoiints RCNN在外在變數(如環境、設備、位置)的改變下，是否影響其擷取人體關鍵點數值的準確性，並進而藉由Keypoints RCNN分析高齡者平衡姿勢的細節，如人體代償機制、姿勢控制策略、肢段的一致性及身體晃動偏移程度，以此評估高齡者平衡功能的品質表現。
在外在變數部份，本研究分別針對隱私及不同的環境光線、設備位置及影像解析度進行測試，測試結果顯示Keypoints RCNN神經網路辨識技術不受以上外在變數影響，主要會造成差異部分為辨識設備的位置，因此針對不同受測者必須採用定位拍攝。於高齡平衡功能評估部分，本研究經由臨床專家協助日照中心兩位高齡長者進行簡易伯格式量表的平衡動作拍攝，採用閉眼維持站姿不扶及單腳站立兩個測試項目，透過Keypoints RCNN擷取其人體共12個關鍵點(不含臉部5個關鍵點)的數據，利用軌跡散佈圖(Scatter Plot)分析兩位受測者關鍵點的移動趨勢，及應用統計特徵值，包括變異係數、標準差及全距量化各關鍵點數據的離散程度，藉此可進一步觀察出高齡者平衡動作的整體移動趨勢、代償表現、關鍵點軌跡的一致性、水平及垂直穩定性，惟於姿勢控制策略，本研究雖可分析出其關鍵點正面相關位移數據，但由於尚欠缺側面拍攝資訊，建議後續研究者可同時於正面及側面進行影像拍攝，應有助於獲得高齡平衡動作更加全面的評估與分析。

The purpose of this study was to investigate the accuracy of capturing human key points on the AI image recognition technology Keyoiints RCNN by changing the external variates (e.g. environment, equipment and location). To study the balance of the elderly postural details, such as human compensatory mechanisms, postural control strategies, consistency of limb segments and degree of body sway offset by using the Keypoints RCNN results.
In the part of external variables, this study conducts experiments on privacy and different ambient light, device location and image resolution. The result shows that the Keypoints RCNN neural network recognition technology is only affected by the location of the identification devices. A well define recording location can apply to different subjects. Therefore, the recognition video of two elderly people was recorded in a long-term care center with the assistance from a clinical expert. Their balance movements of the Short-Form Berg Balance Scale were evaluated in this study. Standing unsupported with eyes closed and stand on one leg are adopted to perform Keypoints RCNN human key points recognition. The vector coordinate values of 12 key points can be obtained (excluding the 5 key points of the face). Use the Scatter Plot to analyze the trajectory trend of the key points of the two subjects, and apply statistical eigenvalues that include the coefficient of variation (C.V.), standard deviation (STD) and range (R) to quantify the discrete degree. These can further observe the overall trajectory trend, compensation performance, the consistency of the key points, horizontal and vertical stability of the balance movement of the elderly. In the posture control strategy, this study can analyze the frontal related displacement value of key points, but due to the lack of lateral information. Therefore, it is suggested that the follow-up researchers can take video from the front and side at the same time, which should help to obtain a more comprehensive evaluation and analysis of balance movements in the elderly.

論文口試委員審定書
中文摘要
Abstract
致謝
目錄
圖目錄
表目錄
第一章 緒論
1.1研究背景
1.2研究動機與目的
1.3研究流程
第二章 文獻探討
2.1 AI影像識別技術相關研究
2.2 簡易伯格式平衡量表(SFBBS)
2.3 人體姿勢平衡的穩定性
2.4 高齡人體關鍵點數據分析
2.4.1 資料視覺化(Data visualization)
2.4.2 統計特徵值的應用
第三章 研究方法
3.1 關鍵點辨識技術的說明
3.2 關鍵點辨識技術應用架構
3.3 外在變數對KEYPOINTS RCNN數據的影響
3.3.1 受測者隱私影響
3.3.2 影像錄製設備位置影響
3.3.3 影像解析度影響
3.3.4 環境明暗度影響
3.4 KEYPOINTS RCNN高齡平衡影片蒐集
3.5 KEYPOINTS RCNN數據於人體姿勢平衡分析說明
第四章 研究結果
4.1 KEYPOINTS RCNN外在變數分析
4.1.1 隱私處理分析結果
4.1.2 錄製設備位置不同分析結果
4.1.3 環境明暗度分析結果
4.1.4 影像解析度分析結果
4.2 KEYPOINTS RCNN高齡平衡分析
4.2.1 測試項目一 閉上眼睛並維持站姿不扶(SFBBS-2)
4.2.2 測試項目二 單腳站立(SFBBS-7)
4.3 研究結果
4.3.1 外在變數對KeypointsRCNN數據擷取之影響
4.3.2 高齡平衡動作分析結果
第五章 結論與未來研究
5.1 研究討論
5.2 結論
5.3 未來研究方向與建議
參考文獻

1. Amiridis, I. G., Hatzitaki, V., & Arabatzi, F. (2003). Age-induced modifications of static postural control in humans. Neuroscience letters, 350(3), 137-140.
2. Aslan, U. B., Cavlak, U., Yagci, N., & Akdag, B. (2008). Balance performance, aging and falling: a comparative study based on a Turkish sample. Archives of gerontology and geriatrics, 46(3), 283-292.
3. B³aszczyk, J. W., & Klonowski, W. o. (2001). Postural stability and fractal dynamics. Acta Neurobiol. Exp, 61, 105-112.
4. Bai, D. L. (2016). 以互動式資料視覺化輔助社區高齡者跌倒風險評估. 福祉科技與服務管理學刊, 4(2).
5. Berg, K., Wood-Dauphinee, S., Williams, J., & Gayton, D. (1989). Measuring balance scale: Reliability assessment with an elderly population. Archives of Physical Medicine and Rehabilitation, 73, 1073-1080.
6. Burton, A. W., & Wade, M. G. (1990). Physical constraints on the movement of disabled performers. In Advances in psychology (Vol. 74, pp. 89-106): Elsevier.
7. Chou, C.-Y., Chien, C.-W., Hsueh, I.-P., Sheu, C.-F., Wang, C.-H., & Hsieh, C.-L. (2006). Developing a short form of the Berg Balance Scale for people with stroke. Physical therapy, 86(2), 195-204.
8. George.wu. (2018). 2017 Top10 CV: 目前最好的object detection模型 Mask R-CNN. Retrieved from http://dosudodl.wordpress.com/2018/02/19/2017-top10-cv-目前最好的object-detection模型-mask-r-cnn/
9. Gill, J., Allum, J., Carpenter, M., Held-Ziolkowska, M., Adkin, A., Honegger, F., & Pierchala, K. (2001). Trunk sway measures of postural stability during clinical balance tests: effects of age. The Journals of Gerontology Series A: Biological Sciences and Medical Sciences, 56(7), M438-M447.
10. Hasson, C. J., Caldwell, G. E., & van Emmerik, R. E. (2008). Changes in muscle and joint coordination in learning to direct forces. Human movement science, 27(4), 590-609.
11. He, K., Gkioxari, G., Dollár, P., & Girshick, R. (2017). Mask r-cnn. Paper presented at the Proceedings of the IEEE international conference on computer vision.
12. Horak, F. B., & Nashner, L. M. (1986). Central programming of postural movements: adaptation to altered support-surface configurations. Journal of neurophysiology, 55(6), 1369-1381.
13. Kim, S.-G., & Kim, M.-K. (2015). The intra-and inter-rater reliabilities of the Short Form Berg Balance Scale in institutionalized elderly people. Journal of physical therapy science, 27(9), 2733-2734.
14. Levin, M. F., Kleim, J. A., & Wolf, S. L. (2009). What do motor “recovery” and “compensation” mean in patients following stroke? Neurorehabilitation and neural repair, 23(4), 313-319.
15. Migut, M., & Worring, M. (2012). Visual exploration of classification models for various data types in risk assessment. Information Visualization, 11(3), 237-251.
16. Patil, C., & Gupta, V. (2021). Human Pose Estimation using Keypoint RCNN in PyTorch. Retrieved from https://learnopencv.com/human-pose-estimation-using-keypoint-rcnn-in-pytorch/
17. Pickering, R. M., Grimbergen, Y. A., Rigney, U., Ashburn, A., Mazibrada, G., Wood, B., . . . Bloem, B. R. (2007). A meta‐analysis of six prospective studies of falling in Parkinson's disease. Movement Disorders, 22(13), 1892-1900.
18. Ren, S., He, K., Girshick, R., & Sun, J. (2015). Faster r-cnn: Towards real-time object detection with region proposal networks. Advances in neural information processing systems, 28, 91-99.
19. Stelmach, G. E., Phillips, J., DiFabio, R. P., & Teasdale, N. (1989). Age, functional postural reflexes, and voluntary sway. Journal of Gerontology, 44(4), B100-B106.
20. Toshev, A., & Szegedy, C. (2014). Deeppose: Human pose estimation via deep neural networks. Paper presented at the Proceedings of the IEEE conference on computer vision and pattern recognition.
21. 朱經明. (2001). 教育統計學: 五南圖書出版股份有限公司.
22. 呂偉僑, 吳淑娟, 蘇希洵, 張敬仁, 孫天龍, & 黃建華. (2015). 以互動式可視化儀表盤做為社區老人跌倒風險評估工具. 台灣復健醫學雜誌, 43(2), 99-110.
23. 國家發展委員會. (2020). 中華民國人口推估（2020至2070年）. Retrieved from https://www.ndc.gov.tw/Content_List.aspx?n=695E69E28C6AC7F3
24. 張嘉玲. (2005). 站立穿褲動作姿勢控制.
25. 梁志光. (2010). 常見老年症候群-跌倒. 高雄醫師會誌, 18(3), 236-241.
26. 陳定邦. (2020). 以顏色變化突顯事件及趨勢之多變量時序資料視覺化.
27. 黃佩珍, 張雅棻, & 游振弘. 結合軀幹侷限與分散式侷限誘發療法於中風後上肢復健之先驅研究.
28. 衛生福利部國民衛生署. (2012). 老人跌倒常見的危險因子有那些? Retrieved from https://www.hpa.gov.tw/Pages/Detail.aspx?nodeid=807&pid=4327
29. 蘇柏霖. (2015). 發展評估靜止站立穩定性的新特徵變數. 中山大學機械與機電工程學系研究所學位論文, 1-101.