Basic Search / Detailed Display

Author: 謝孟寰
Hsieh, Meng-Huan
Thesis Title: 基於人工智慧的功能性動作檢測
An Automatic FMS Method Based on AI Technic
Advisor: 吳順德
Degree: 碩士
Master
Department: 機電工程學系
Department of Mechatronic Engineering
Thesis Publication Year: 2020
Academic Year: 108
Language: 中文
Number of pages: 55
Keywords (in Chinese): 人工智慧肢體偵測物理治療物理治療
Keywords (in English): AI, Limb detection, Physiotherapy, Rehabilitation
DOI URL: http://doi.org/10.6345/NTNU202001381
Thesis Type: Academic thesis/ dissertation
Reference times: Clicks: 225Downloads: 67
Share:
School Collection Retrieve National Library Collection Retrieve Error Report
  • 運動傷害一直是運動員最害怕的事情,無論是慢性運動傷害、急性運動傷害都可能縮短甚至終結一個人的運動生涯,所以運動傷害防護在運動界與醫療界一直是備受關注的重點之一,經過專業人士多年研究,已經能夠預防與治療運動傷害,也衍伸專門做運動防護與治療的職業,例如運動防護員、物理治療師、運動管理師…等。
    但隨著運動風氣興起現今已是全民運動的時代,並非所有運動團隊都有能力聘請一位防護員,自主運動的民眾往往沒辦法做到預防,只能在身體開始有傷痛再就醫,何況在全球仍有許多醫療資源較不方便的地區,當地民眾要花費更高的時間與金錢成本才能到達最近的復健科或是物理治療工作室。
    隨著人工智慧(AI)蓬勃發展,AI的運算速度與準確性不斷被改善,現今已經能使用二維影像訊號進行人體姿態識別,一旦能用影像計算出人類肢體姿態,自動檢測肢體健康狀況與輔助診斷將成為未來發展方向,除了能降低專業人力資源的依賴還能省去傳統物理治療中人工測量角度的麻煩。
    本研究以卡內基梅隆大學開發的Openpose肢體偵測AI模型為基礎,參考美國國家運動醫學學會(NASM)出版的「矯正運動訓練要素」、功能性動作檢測為診斷依據,結合物理治療師與復健科醫師臨床知識,開發出2D影像的肢體健康檢測系統。系統分為手機App使用者操作介面、運算伺服器、專業人士介面與資料庫等三個主要部分,由APP錄製的使用者影片,運算伺服器先經Openpose計算人體關鍵點(Keypoint)再轉換成臨床檢測指標,最終將健康指標與動作評分顯示在App與專業人士介面,給予使用者健康度回饋並推薦適合的復健影片。運算伺服器分為肢體偵測與檢測系統兩大部分,其中肢體偵測細分樣本採集、影片處理、關鍵點(Keypoint)處理、結果比較四個部份,檢測系統細分為系統架構、健康評等、影片推薦判斷依據。

    Sports injuries have been the most feared thing for athletes. Both chronic and acute sports injuries might shorten or even end one’s sports career. Therefore, it has always been one of the focuses of professionals in sports and medical fields. After years of research by professionals, there have been ways to prevent and treat sports injuries. It also developed occupations specialized in sports protection and treatment, such as sports protectors, physical therapists, and sports managers.
    With the rise of sports culture, sport is national activity now. However, not all sports teams have the ability to hire a protector, not to mention the people who exercise independently. They have difficulties to prevent it, but have medical treatments after sports injuries. Needless to say that medical resources are inconvenient in some areas. Going to the nearest rehabilitation department or physical therapy studio may take lots of time in these areas.
    With the rapid development of artificial intelligence (AI), the computing speed and accuracy of AI are continuously optimized. Nowadays, computers can be used for human body recognition. Once the positions of body pose can be identified with image detection, automatic detection of sports injuries and aided diagnosis will become the development direction in the future. It can not only reduce the dependence on professional resources, but also abandon the need for manual angle measurement in traditional physical therapy.
    This research is based on the Openpose pose detection AI model developed by Carnegie Mellon University. To make "Essentials of Corrective Exercise Training" published by the National Academy of Sports Medicine (NASM) and functional motion detection as the basis of judgment, this research combined the clinical judgment knowledge from physical therapists and rehabilitation physicians and developed a 2D image detection system for sports injuries. Mobile phone App user interface, computing server, and professional interface and database are three main parts in this system. Users will receive their health feedback and be recommend suitable rehabilitation videos from the system after it get users’ films. This research includes pose detection and analysis system. Pose detection consists of sample collection, video processing, keypoint processing, and result comparison. The detection system is comprised of system architecture, health evaluation, and the basis of video recommendation judgment.

    摘要 i Abstract iii 誌謝 v 目錄 viii 表目錄 x 圖目錄 xi 第一章、緒論 1 1.1 前言 1 1.2 研究動機與目標 2 1.3 論文章節概述 5 1.4 文獻探討 8 第二章、肢體健康檢測系統 14 2.1 使用者App 15 2.2 資料庫 21 2.3 運算伺服器 22 2.4 專業人士軟體 25 第三章、樣本採集 28 3.1 樣本動作 28 3.2 拍攝要點與受測樣本 32 第四章、實驗設計與結果討論 34 4.1 影片前處理 34 4.2 關鍵點處理 36 4.3 結果比較 42 4.4 結果與討論 43 第五章、結論與未來展望 51 5.1 結論 51 5.2 未來展望 52 參考文獻 54

    [1] CMU-Perceptual-Computing-Lab, Openpose, GitHub, [Online], Available: https://github.com/CMU-Perceptual-Computing-Lab/openpose
    [2] Z. Zhang, “Microsoft Kinect Sensor and Its Effect”, IEEE Xplore, Volume: 19 Issue: 2, pp. 4-10, Feb. 2012
    [3] S.-C. Yang, “Human Action Recognition Using Kinect Camera”, ISU, master, 2013
    [4] Sportzorgmasseurnet, Sportzorgmasseur volwaardige samenwerkings-partner in de sportzorgketen, Sportzorgmasseurnet, [Online], Available: https://www.sportzorgmasseurnet.nl/, May. 2016
    [5] 黃博靖, 關節活動度測量:角度測量學習指引(Measurement of Joint Motion: A Guide to Goniometry ), 合記圖書出版社, 2018
    [6] D.S. Teyhen, S.W. Shaffer, C.L. Lorenson, J.P. Halfpap, D.F. Donofry, M.J. Walker, J.L. Dugan, J.D. Childs, “The Functional Movement Screen: A Reliability Study”, Jospt, Volume:42 Issue:6, pp. 530-540, Jun. 2012
    [7] M.A. Clark, B.G. Sutton, S.C. Licett, “NASM Essential of Corrective Exercise Training”, Jones & Bartlett Learning, Jun. 2013
    [8] Z. Cao, G. Hidalgo, T. Simon, S.-E. Wei, Y. Sheikh, “OpenPose: realtime multi-person 2D pose estimation using Part Affinity Fields”, arxiv:1812.08008, Dec. 2018
    [9] H. Joo, T. Simon, X. Li, H. Liu, L. Tan, L. Gui, S. Banerjee, T. Godisart, B. Nabbe, I. Matthews, T. Kanade, S. Nobuhara, Y. Sheikh, “Panoptic Studio: A Massively Multiview System for Social Interaction Capture”, IEEE Xplore, Volume: 41 Issue: 1 , pp. 190 - 204, Jan. 2019
    [10] Z. Cao, T. Simon, S.-E. Wei, Y. Sheikh, “OpenPose: Realtime Multi-Person 2D Pose Estimation using Part Affinity Fields”, arXiv:1812.08008, Dec. 2018
    [11] 葉峻榳, 功能性動作檢測 Functional Movement Screen (FMS) , [Online], Available: https://chunting.me/functional-movement-screen-fms/
    [12] S. Qiao, Y. Wang, J. Li, “Real-time human gesture grading based on OpenPose”, IEEE Xplore, Oct. 2017
    [13] 薛漪平, 生理疾病職能治療學:I評估理論與技巧, 禾楓書局, 2011

    下載圖示
    QR CODE