人的走路姿態會呈現生理資訊，因此步態分析方法可以用來評估一個人的身體狀態。而有很多步態分析的方法都需要配戴感知器來記錄受測者的走路行為，我們提出一種不需要配戴感知器的方法，它是透過影像來取得走路特徵資訊，並使用這些走路特徵資訊來進行步態分析。我們的方法會對影像做人體分割，從分割出來的腿部得到走路特徵資訊。
藉由肺功能機，得知受測者的身體狀態，其呼吸功能參數則為評估受測者身體狀況的標準。如果我們的分類結果與肺功能機分類結果相似，亦或是預測出來的數值與原本的呼吸功能參數相近，則可以證明我們的分析方法與肺功能機具有關連性，進而成功得知受測者當下的身體狀態。
我們在新北市雙和醫院拍攝受測者進行六分鐘快走測試的過程，同時取得這些受測者藉由肺功能機得到的呼吸功能參數，根據這些呼吸功能參數將受測者分為等級一、等級二、等級三。
根據走路特徵資訊，我們對受測者進行分類與預測呼吸參數的實驗。在受測者分類的實驗中，將等級一與等級二的受測者分為差的組別，而等級三的受測者分為好的組別，我們的分類結果有75%的準確率。在預測呼吸參數的實驗中，我們在預測FEV1與FVC上有0.69與0.67的關聯性，而在預測FEV1/FVC時只有0.25的關聯性，顯示我們的分析方法在預測單一參數時有較高的關聯性。從分類受測者實驗與預測呼吸參數實驗中，可以證明我們的分析方法與肺功能機有良好的關聯性。
此外，使用我們的走路特徵資訊加上雷達系統的特徵資訊，其分類結果會從75%上升到81%，而在預測FEV1/FVC時，關聯性也從0.25上升到0.42，因此與雷達系統的結合可以增強我們的分析方法與肺功能機的關聯性

The human walking behavior could express the physiological information of human body. Consequently, gait analysis methods can be used to access the human body condition. However, many gait analysis methods need to wear sensors to record the walking behavior. We propose a vision sensor based gait analysis method without wearing any sensors. Our method would segment the silhouette in order to extract legs parts. From legs part, we could extract the gait features.
The subjects can get the respiratory parameters by using pulmonary spirometer. These parameters become the standard of accessing the body condition of the subjects. If the classification results of our experiment are similar to the classification results by respiratory parameters or the predicting values are close to the respiratory parameters, we can prove there is a correlation between pulmonary spirometer and our method and understand the body condition of the subjects.
We film the subjects when they are running the six-minutes brisk walking test at Shuang-Ho hospital in New Taipei, Taiwan and get the respiratory parameter by using pulmonary spirometer. These subjects would be divided into level 1, level 2, and level 3 depending on the respiratory parameters.
We would run classification and predicting experiment according to our extracted features. In the classification experiment, the subjects of level 1 and level 2 belong to Bad group and the subjects of level 3 belong to Good group. In the classification experiment, the accuracy result is 75%. In predicting experiment, the correlations of predicting FEV1 and FVC are 0.69 and 0.67. However, the correlation of predicting FEV1/FVC is 0.25. The results have higher correlation on predicting the parameters. As a result, there is a correlation between the pulmonary spirometer and our method.
Combining the features of radar system with our features, the classification result would improve to 81% form 75 %. In predicting FEV1/FVC, the correlation also improves to 42% from 25%. Therefore, cooperating with radar system would improve the correlation of pulmonary spirometer and our method.

1 Introduction 1
1.1 Motivation -- 1
1.2 Problem Description -- 2
1.3 Goal and Contribution -- 3
1.4 Thesis Organization -- 4
2 Related Works 5
2.1 COPD studies -- 5
2.2 Gait analysis -- 6
3 Vision Sensor based Gait Analysis 8
3.1 Proposed flow -- 8
3.2 Pre-processing -- 10
3.2.1 Background subtraction -- 10
3.2.2 Shadow removal -- 10
3.2.3 Connected Component Labeling (CCL) -- 13
3.3 Feature Extraction -- 13
3.3.1 Segmentation -- 15
3.3.2 Feature Extraction -- 17
3.4 Gait Analysis -- 17
4 Proposed Gait features 22
5 Clinical Experiment Environment 26
5.1 Experiment set up and flow -- 26
5.2 Data Collection -- 27
6 Experimental Results 34
6.1 Classification with Support Vector Machine -- 35
6.2 Prediction with Adaptive Neural Fuzzy Inference Systems -- 41
6.3 Cooperating with Radar system -- 45
7 Conclusion and Future Work 52
7.1 Conclusion -- 52
7.2 Future Work -- 53
.1 The SVM classification results with different inputs -- 58

[1] C. C. Foundation, "http://www.clevelandclinicmeded.com/medicalpubs/
diseasemanagement/pulmonary/pulmonary-function-testing/," 2003.
[2] I. Yunfeng Wu, Member and I. Sridhar Krishnan, Senior Member, "Sta-tistical analysis of gait rhythm in patients with parkinsons disease," IEEE TRANSACTIONS ON NEURAL SYSTEMS AND REHABILITATION ENGINEERING, no. 2, pp. 150-158, APRIL 2010.
[3] M.-F. Wu, C.-L. Chen, C.-Y. Wen, and J.-Y. Hsu, "Design of pervasive rehabilitation monitoring for chronic obstructive pulmonary disease," IEEE SENSORS JOURNAL, no. 11, pp. 4413-4422, NOVEMBER 2013.
[4] I. Adam M. Howell, Student Member, T. Kobayashi, H. A. Hayes, K. B. Foreman, and I. Stacy J. Morris Bamberg, Senior Member, "Kinetic gait analysis using a low-cost insole," IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, no. 12, pp. 3284-32 900, DECEMBER 2013.
[5] F. LM and H. SS, "Global strategy for the diagnosis, management and prevention of copd: 2003 update," European Respiratory Journal, vol. 22, no. 1, pp. 1-1, 2003.
[6] W. H. Organization, The world health report 2000, 2000.
[7] "Mortality by cause for eight regions of the world: Global burden of disease study."
[8] A. T. F. on Pulmonary Rehabilitation, "An ocial european respiratory society statement on physical activity in copd," European Respiratory Journal, December 2014.
[9] S. D. Sullivan, S. D. Ramsey, and T. A. Lee, "The economic burden of copd," Chest. 2000, pp. 5S-9S, February 2000.
[10] V. SN, K. HS, T. T, and L. JW, "Level of daily physical activity in individuals with copd compared with healthy controls," Respir Res, March 2011.
[11] E. G.-S. et al., \Determinants and outcomes of physical activity in patients
with copd: a systematic review," Thorax, February 2014.
[12] P. Shu-Yi Liao, MD, M. Roberto Benzo, M. Andrew L. Ries, MD, and P. Xavier Soler, MD, "Physical activity monitoring in patients with chronic obstructive pulmonary disease," Journal of the COPD Foundation, pp. 155-165, August 2014.
[13] M. Merone, L. Onofri, P. Soda, C. Pedone, and R. A. I. G. Iannello, "Early experiences in copd exacerbation detection," ser. 27. International Symposium on Computerased Medical Systems, 11 2014, pp. 606-610.
[14] C. Bellos, A. Papadopoulos, R. Rosso, and D. I. Fotiadis, "Identication of copd patients' health status using an intelligent system in the chronious wearable platform." IEEE J. Biomedical and Health Informatics, vol. 18, no. 3, pp. 731-738, 2014.
[15] L. Wen, J. Qian, X. Hu, L. Shen, X. Wu, and C. Yu, "Gait measurement and quantitative analysis in patients with parkinson's disease for rehabilitation assessment." 2013, pp. 286-291.
[16] P.-C. Chung, Y.-L. Hsu, C.-Y. Wang, C.-W. Lin, J.-S. Wang, and M.-C. Pai, "Gait analysis for patients with alzheimer's disease using a triaxial accelerometer." 2012, pp. 1323-1326.
[17] S. Jiang, B. Zhang, and D. Zou, Guobing andWei, "The possibility of normal gait analysis based on a smart phone for healthcare." 2013, pp. 2235-2240.
[18] J. Schiel and R. Green, "Adaptive human silhouette extraction with chromatic distortion and contour tracking," ser. 28. International Conference on Image and Vision Computing, 11 2013.
[19] V. N. Vapnik, The nature of statistical learning theory. New York: Springer, 1995.
[20] R. Jang and shing Jyh, "Ans: Adaptive-network-based fuzzy inference system," 5 1993.
[21] A. Al-Hmouz, J. Shen, R. Al-Hmouz, and J. Yan, "Modeling and simulation of an adaptive neuro-fuzzy inference system (ans) for mobile learning." IEEE
TRANSACTIONS ON LEARNING TECHNOLOGIES, vol. 5, no. 3, pp.
226-237, 2012.
[22] a. Y. Y. T. Jie Xu, B. Zou, Z. Xu, and a. Y. L. Luoqing Li, "The generalization ability of online svm classication based on markov sampling," IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS, vol. 26, no. 3, pp. 628-639, 5 2015.