基於深度學習之復健動作辨識系統｜國立中央大學博碩士論文系統

簡易檢索 / 詳目顯示

回結果列表

研究生：	戴邦地 Pang-Ti Tai
論文名稱：	基於深度學習之復健動作辨識系統 A Deep Learning Approach to the Recognition of Rehabilitation Exercises
指導教授：	蘇木春
口試委員:
學位類別：	碩士 Master
系所名稱：	資訊電機學院 - 資訊工程學系 Department of Computer Science & Information Engineering
論文出版年：	2017
畢業學年度：	105
語文別：	中文
論文頁數：	79
中文關鍵詞：	深度學習、動作辨識、Kinect 2 、復健
外文關鍵詞：	deep learning, motion recognition, Kinect 2, rehabilitation
相關次數：	點閱：14 下載：0
分享至:	分享至facebook 分享至twitter

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

　　現代社會有許多人因各種疾病傷痛導致生活機能受損，為了幫助患者回復受損的機能，復健醫療是不可或缺的。然而常常因為患者眾多，而面臨醫療人力與醫療資源不足的問題。此時若能將復健醫療結合現代科技發展輔助系統，讓患者於家中也能從事適當且規律的復健運動，將有助於改善患者受損的生活機能。
　　復健輔助系統須能正確辨識患者的動作，才能有效進行復健運動，此時動作辨識的準確率扮演了十分重要的角色。本論文結合Kinect 2感應器與深度學習技術，提出一套動作辨識系統。對於動作的特徵擷取，本論文基於人的深度影像提出一套新的方法，此方法的主要概念是將與時間相關的一連串動作轉換成一張軌跡圖表示，本論文稱之為動作軌跡圖。以動作軌跡圖作為動作的特徵，利用卷積神經網路進行動作辨識。
　　本論文以12種復健動作為例進行實驗，在動作軌跡圖方法的各個步驟使用不同方法進行測試與比較，最後採用效果較好的方法。實驗結果顯示本論文提出之動作軌跡圖方法結合深度學習技術在動作辨識上具有不錯的辨識能力。

　　Many people suffer from inconveniences due to various kinds of diseases or bodily injury in modern society. In order to help these people, physical rehabilitation is indispensable. However, too many patients may cause a shortage of medical resource. Therefore, if a combination of physical rehabilitation and modern technology could be made, patients can improve their health condition by performing proper rehabilitation exercises in their own place with a rehabilitation system.
　　A rehabilitation system should be able to recognize the motion performed by patient correctly. The accuracy of motion recognition plays a very important role in a rehabilitation system. This paper provides a motion recognition system which uses the Kinect 2 sensor with the deep learning techniques. This paper introduces a new method of feature extraction. The main concept of this method is to convert a motion into a motion trajectory image. The motion trajectory image is then used as the input of convolutional neural networks for motion recognition.
　　This paper uses 12 types of rehabilitation exercises in our experiment. We have tried different ways in each step of our method, and we finally choose one with a better result in our test. According to the result, our method has a good ability in motion recognition.

摘要    i
ABSTRACT    ii
誌謝    iii
目錄    iv
圖目錄    vi
表目錄    vii
第一章、緒論    1
1-1 研究動機    1
1-2 研究目的    2
1-3 論文架構    3
第二章、相關研究    4
2-1 復健科技    4
2-2 動作辨識    8
2-2-1 人體骨架動作辨識    8
2-2-2 影像動作辨識    9
2-3 分群演算法    10
2-3-1 分群使用之距離計算方法    10
2-3-2 k-means演算法    11
2-3-3 適應共振理論    12
2-4 降維演算法    13
2-4-1 Sammon投影    13
2-4-2 用於SOM之降維演算法    13
2-5 類神經網路    16
2-5-1 類神經網路簡介    16
2-5-2 卷積神經網路    18
2-5-3 深度學習工具TensorFlow    21
第三章、研究方法    23
3-1 流程介紹    23
3-2 人像擷取    24
3-3 特徵擷取    25
3-3-1 Motion Energy Image    25
3-3-2 Motion History Image    26
3-3-3 動作軌跡圖    27
3-4 基本動作分群    29
3-4-1 k-means演算法    29
3-4-2 適應共振理論    30
3-5 動作軌跡圖產生    31
3-5-1 Sammon投影    31
3-5-2 用於SOM之降維演算法    32
3-5-3 投影結果    32
3-5-4 軌跡作圖    33
3-6 分類器    36
第四章、實驗設計與結果    37
4-1 實驗設計    37
4-1-1 實驗環境架設    37
4-1-2 實驗動作設計    40
4-1-3 K次交叉驗證    40
4-1-4 實驗步驟    41
4-2 實驗結果    44
4-2-1 基本動作分群輸入資料之比較結果    44
4-2-2 不同距離計算方法之比較結果    47
4-2-3 不同分群方法之比較結果    49
4-2-4 不同降維方法之比較結果    52
4-2-5 動作軌跡圖方法實驗結果    55
4-2-6 對人的推廣性之驗證結果    57
4-2-7 不同特徵擷取方法之比較結果    59
第五章、結論與未來展望    61
5-1 結論    61
5-2 未來展望    62
參考文獻    63
附錄一、12種復健動作    66
                                

[1] 復健 - 維基百科. [Online]. Available: https://zh.wikipedia.org/wiki/復健. [Accessed: 17-May-2017].
[2] 創新復健！引進新儀器增加病患復健意願－TVBS新聞網. [Online]. Available: https://news.tvbs.com.tw/health/677527. [Accessed: 17-May-2017].
[3] 遠見雜誌：萬華醫院推動「一站式復健服務」. [Online]. Available: https://www.gvm.com.tw/webonly_content_11482.html. [Accessed: 17-May-2017].
[4] Doctor Kinetic. [Online]. Available: http://doctorkinetic.com/. [Accessed: 17-May-2017].
[5] 臺北榮民總醫院「智慧醫療復健系統」. [Online]. Available: https://www.vghtpe.gov.tw/News!one.action?nid=2986. [Accessed: 17-May-2017].
[6] 希望之手－Rehab-Robotics. [Online]. Available: http://www.rehab-robotics.com/zh-hk/hoh/index.html. [Accessed: 17-May-2017].
[7] G. D. Sosa, J. Sánchez, and H. Franco, “Improved front-view tracking of human skeleton from Kinect data for rehabilitation support in Multiple Sclerosis,” 20th Symposium on Signal Processing, Images and Computer Vision, 2015.
[8] H. Tanaka, I. Kajitani, K. Homma, Y. Wakita, and Y. Matsumoto, “A Motion Tracker Using High-Accuracy AR Markers for On-site Motion Analysis,” IEEE International Conference on Systems, Man, and Cybernetics, pp. 1427-1432, 2015.
[9] S. Karungaru, “Human Action Recognition using Wearable Sensors and Neural Networks,” 10th Asian Control Conference, 2015.
[10] M. Singh, A. Basu, and M. Kr. Mandal, “Human Activity Recognition Based on Silhouette Directionality,” IEEE Transactions on Circuits and Systems for Video Technology, Vol. 18, No. 9, pp. 1280-1292, 2008.
[11] B. Jagadeesh and C. M Patil, “Video based action detection and recognition human using optical flow and SVM classifier,” IEEE International Conference on Recent Trends in Electronics, Information & Communication Technology, pp. 1761-1765, 2016.
[12] A. Bobick and J. Davis, “Real-time recognition of activity using temporal templates,” Applications of Computer Vision WACV '96, pp. 39-42, 1996.
[13] Pearson correlation coefficient - Wikipedia. [Online]. Available: https://en.wikipedia.org/wiki/Pearson_correlation_coefficient. [Accessed: 20-May-2017].
[14] Euclidean distance – Wikipedia. [Online]. Available: https://en.wikipedia.
org/wiki/Euclidean_distance. [Accessed: 20-May-2017].
[15] 蘇木春、張孝德，機器學習：類神經網路、模糊系統以及基因演算法則，三版，全華圖書股份有限公司，臺北市，民國九十三年。
[16] Adaptive resonance theory – Wikipedia. [Online]. Available: https://en.wikipedia.org/wiki/Adaptive_resonance_theory. [Accessed: 20-May-2017].
[17] J. W. Sammon, “A Nonlinear Mapping for Data Structure Analysis,” IEEE Transactions on Computers, Vol. C-18, No. 5, pp. 401-409, 1969.
[18] M. C. Su and H. T. Chang, “Fast Self-Organizing Feature Map Algorithm,” IEEE Transactions on Neural Networks, Vol. 11, No. 3, pp. 721-733, 2000.
[19] Convolutional neural network – Wikipedia. [Online]. Available: https://en.wikipedia.org/wiki/Convolutional_neural_network. [Accessed: 22-May-2017].
[20] Convolution Artificial Inteligence. [Online]. Available: https://leonardoaraujosantos.gitbooks.io/artificial-inteligence/content/
convolution.html. [Accessed: 22-May-2017].
[21] TensorFlow - Wikipedia. [Online]. Available: https://zh.wikipedia.org/
wiki/TensorFlow. [Accessed: 23-May-2017].
[22] Deep MNIST for Experts - TensorFlow. [Online]. Available: https://www.tensorflow.org/get_started/mnist/pros. [Accessed: 23-May-2017].
[23] Kinect 硬體. [Online]. Available: https://developer.microsoft.com/zh-tw/windows/kinect/hardware. [Accessed: 24-May-2017].
[24] Kinect for Windows SDK v2 基本介紹 | Heresy's Space. [Online]. Available: https://kheresy.wordpress.com/2014/12/29/kinect-for-windows-sdk-v2-basic/. [Accessed: 24-May-2017].
[25] JointType Enumeration. [Online]. Available: https://msdn.microsoft.com/
en-us/library/microsoft.kinect.jointtype.aspx. [Accessed: 24-May-2017].
[26] 張智傑，「帕金森氏症病人復健輔助系統」，國立中央大學，碩士論文，民國101年。
[27] Cross-validation (statistics) – Wikipedia. [Online]. Available: https://en.wikipedia.org/wiki/Cross-validation_(statistics). [Accessed: 25-May-2017].

簡易檢索 / 詳目顯示

相關論文