在許多使用影像分析人類行為的人機互動研究中，常常可以看到為了一個良好的辨識率，往往需要透過限制使用者與感測器的距離和角度，使得在平常的應用中是不方便或是不可行的。為了使基於影像感測技術的辨識應用能有效率於平常居家生活中使用，所發展辨識系統就必須考慮使用者在不熟悉系統或特別是針對不良於行的系統使用者之無法輕易移動至系統所需的最佳辨識位置的問題。為克服在進行系統辨識時之使用者與影像感測器之間的方位問題(不當的距離和角度之感測器方位將影響辨識率)，本論文提出一個貝氏網路模型方法，此方法可在三維感測影像辨識應用中達到最佳感測器的自動定位，即使當使用者與感測器的相對方位不理想時，本論文所提的方法仍能透過感測三維人體影像特徵而進一步計算出使用者與影像感测裝置的最佳的相對位置。本論文所提出的影像感測自動定位方法將能提供給現今的機器人移動裝置進行感測定位整合。
本論文所提定位方法之使用於姿勢或手勢的感測辨識應用中，主要是先藉由三維(3D)影像感測器擷取的人體骨架座標，這些座標將依照不同辨識應用的而轉換為位置定位特徵。在姿勢辨識應用的感測定位中，「方位」及「距離」等兩個類型的屬性特徵共有七組，在手勢辨識應用的感測定位中，「方位」、「距離」及「手部」等三個類型的屬性特徵共有九組。本論文所提出的感測自動定位為一種有效結合隱藏式馬可夫模型(HMM)的貝氏網路(BN)方法(3DBN-HMM)，本方法將有效考量HMM的辨識決策結果及模型相似分數而進一步建立 3DBN-HMM的影像定位估算模型。本論文在建立3DBN-HMM貝氏網路模型時同時考量了網路參數訓練方式及網路結構訓練等兩個模型訓練方式。所建立的3DBN-HMM模型實施於影像辨識的最佳位置的決策時，3DBN-HMM 節點的事件是藉由 HMM 的辨識結果決定節點的事件發生，使用者的定位完成後，將透過定位的結果將三維影像感測器移動至最佳的辨識位置，以維持基於三維影像感測的辨識的性能在一個水準之上。
在本論文所提出之最佳感測影像定位的模型學習資料，是藉由數組 HMM 的辨識結果及其相似分數進行 3DBN-HMM 的學習，用於姿勢辨識應用定位的 HMM 的平均辨識性能為63.15%，用於手勢辨識應用定位的HMM 的平均辨識性能為59.57%，使用進行學習後的3DBN-HMM模型進行姿勢辨識定位及手勢辨識定位的位置決策的準確度皆約為70%，此位置準確度已較目前Weka 軟體所提供的貝氏模型的定位準確度提升約10%的效能。在使用3DBN-HMM定位方法的姿勢辨識及手勢辨識的動作決策的準確度皆表現良好，對於不熟悉系統或無法輕易移動至系統的不良於行的系統使用者而言，使用3DBN-HMM定位方法的姿勢辨識的辨識性能平均達86.11%，使用3DBN-HMM定位方法的手勢辨識辨識性能平均達78.83%，使用本論文所提出之定位方法的兩個三維影像辨識應用皆已接近可自行完成辨識定位的一般使用者的手勢或姿勢動作的辨識準確率。

It is often possible seen to have a good recognition rate on many human-computer interaction studies focused on image analysis of human behavior. In these seen studies, the distance and angle of the user are often required to be limited through the sensor so that the usual application will become inconvenient or not feasible. In order to make the image sensing-based application to be conveniently in the home life, it is necessary to consider the best recognition position of the image sensor for those users that cannot easily move to the system (or those users that are unfamiliar with the system). Therefore, it is necessary to overcome the problem of the distance and angle of the sensor to the user to keep the fine recognition rate. In this thesis, even when the user position to the sensor is in an unfavorable recognition position, the user gesture is still finely identified through the designed feature derived from three-dimensional image human body data.
In this thesis, when doing recognition, the human skeleton coordinates captured by the 3D image sensor are converted into azimuth attribute features, distance attribute features, and hand attribute features according to two different application positions. In optimal positioning of gesture recognition applications, there are seven different feature classes including azimuth attribute and distance attribute features. In optimal positioning of hand gesture recognition applications, there are nine different feature including orientation attribute, distance attribute and hand attribute characteristics features. In addition, HMM is used for decisions of event probabilities of Bayesian network (BN) nodes. The BN network parameter learning and BN network structure learning of the 3DBN-HMM are carried out through the HMM decision results and the computed similar scores. In the 3DBN-HMM recognition decision, the event appearance probability of each 3DBN-HMM node is determined by the identification result of the HMM. After optimal positioning of the user is completed by 3DBN-HMM, the 3D image sensor will be able to move to the best position to the user through the determined position result. 3DBN-HMM position recognition in this thesis is maintained at a good recognition performance level for 3D image sensing applications.
In the 3DBN-HMM learning data of this thesis, the 3DBN-HMM is learned by the identification result of the HMM set and all its model similarity scores. The average recognition performances of the HMM for the gesture recognition application and hand gesture recognition application are 63.15% and 59.57%. The accuracy of the built 3DBN-HMM model on position recognition for two gesture recognition applications is about 70%, which is higher than the popular Weka software BN model on 10%. Both of gesture and hand gesture recognition applications by 3DBN-HMM can maintain good recognition performance (86.11% on gesture recognition applications and 78.83% on hand gesture recognition applications).

摘要..............i
Abstract..............iii
誌謝..............v
目錄..............vi
表目錄..............viii
圖目錄..............x
第一章 緒論..............1
1.1 研究動機..............1
1.2 文獻回顧與探討..............2
1.3研究方向..............3
1.4章節概要..............3
第二章 背景技術介紹..............4
2.1KINECT人體姿態特徵擷取..............4
2.2 貝氏網路推論..............7
2.2.1 貝氏網路推論[57]..............7
2.2.2 貝氏網路..............8
2.2.3 吉布斯取樣(Gibbs Sampling)[60]..............10
2.3隱藏馬可夫模型(Hidden Markov model, HMM)..............11
2.3.1Viterbi演算法..............12
2.4三維感測影像應用概述..............13
2.4.1基於三維感測器之姿勢辨識應用..............13
2.4.2基於三維感測器之姿勢辨識應用..............14
第三章 貝氏網路模型機制於三維感測影像應用之最佳定位系統..............16
3.1服務系統定位的三維感測影像之特徵擷取..............18
3.1.1關節位置類別特徵..............20
3.1.2 相鄰關節長度類別特徵..............21
3.1.3 面積類別特徵..............22
3.1.4 角度類別特徵..............23
3.1.5手部類別特徵..............24
3.1.6 3D 影像特徵參數設計總結..............25
3.2應用於貝氏網路節點的三維感測影像特徵..............27
3.3 3DBN-HMM的3D影像定位的模型訓練方法..............29
3.3.1影像定位用之 HMM 模型建立與辨識..............29
3.3.2貝氏網路模型之參數學習..............35
3.3.3 3DBN-HMM影像定位模型的貝氏網路結構學習..............36
3.4 3DBN-HMM 的 3D 影像定位的辨識決策方法..............38
第四章 實驗設計與位置Database錄製..............40
4.1實驗環境設定與資料庫建置..............41
4.2 基於三維影像的使用者定位實驗設定..............53
4.2.1 運用於 3DBN-HMM 的各個 HMM 模型辨識實驗..............54
4.2.2 3DBN-HMM 的預測模型分析..............58
4.2.3 用於三維影像感測的 3DBN-HMM 定位辨識實驗..............62
4.3 基於三維影像最佳定位的三維影像感測辨識實驗設定..............63
4.3.1 基於最佳辨識位置修正的姿勢辨識實驗..............64
4.3.2 基於最佳辨識位置修正的手勢辨識實驗..............68
第五章 結論..............72

[1]蔡祈賢，「少子高齡化社會與人力資源管理之預應」，行政院人事行政總處，2011。
[2]史威德，「陪伴機器人之基於情感辨識音樂播放器系統」，國立交通大學電機資訊國際學位學程碩士論文，2012。
[3]林恩州，「具表情互動之智慧型照護機器人研究開發」，長庚大學醫療機電工程研究所碩士論文，2007。
[4]高聖涵，「肌電圖回授復健機器人於中風病患踝關節主動扭矩控制之研究」，國立成功大學機械工程學系碩士論文，2009。
[5]C. Chen, K. Liu, R. Jafari and N. Kehtarnavaz, “Home-based senior fitness test measurement system using collaborative inertial and depth sensors,” 2014 36th Annual International Conference of the IEEE Engineering in Medicine and Biology Society, pp. 4135-4138, 2014.
[6]A. K. Mishra, M. Skubic and C. Abbott, “Development and preliminary validation of an interactive remote physical therapy system,” 2015 37th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC), pp. 190-193, 2015.
[7]M. M. Kılıc, O. C. Muratlı and C. Catal, ”Virtual reality based rehabilitation system for Parkinson and multiple sclerosis patients,” 2017 International Conference on Computer Science and Engineering (UBMK), pp. 328-331, 2017.
[8]D. Ferreira, R. Oliveira and O. Postolache, “Physical rehabilitation based on kinect serious games,” 2017 Eleventh International Conference on Sensing Technology (ICST), pp. 1-6, 2017.
[9]N. Anbarsanti and A. S. Prihatmanto, “HMM-based model for dance motions with pose representation,” 2014 IEEE 4th International Conference on System Engineering and Technology (ICSET), pp. 1-7, 2014.
[10]M. Z. Uddin, J. Torresen and T. Jabid, “Human activity recognition using depth body part histograms and hidden Markov models,” 2016 International Conference on Innovations in Science, Engineering and Technology (ICISET), pp. 1-4, 2016.
[11]張哲維，「運用3D運動特徵於人體姿態辨識及其學習方法之研究」，國立虎尾科技大學電機工程學系碩士論文，2014。
[12]吳宗桂，「運用KINECT姿態辨識的使用者辨識研究」，國立虎尾科技大學電機工程學系碩士論文，2015。
[13]C. Zou and Z. Liu, “Behavior classification method based on skeleton model from video images,” 2008 International Conference on Computer Science and Information Technology, pp. 309-312, 2008.
[14]Q. Xiao, H. Xiaoxia and S. Gao, “Multi-view 3D sense retrieval approach based on dynamic Bayesian networks,” 2011 IEEE 3rd International Conference on Communication Software and Networks, pp. 212-215, 2011.
[15]T. F. Ansary, M. Daoudi and J. P. Vandeborre, “A bayesian 3-D dearch engine using adaptive views clustering,” IEEE Transactions on Multimedia, pp. 78-88, 2007.
[16]D. R. Faria, M. Vieira, C. Premebida and U. Nunes, “Probabilistic human daily activity recognition towards robot-assisted living,” 2015 24th IEEE International Symposium on Robot and Human Interactive Communication (RO-MAN), pp. 582-587, 2015.
[17]J. Li and Y. Zhang, “A method for learning bayesian network dtructure,” 2014 Sixth International Conference on Intelligent Human-Machine Systems and Cybernetics, pp. 222-225, 2014.
[18]R. L. Trejo, J. P. G. Vázquez, M. L. G. Ramirez, L. E. V. Corral and I. R. Marquez, “Hand goniometric measurements using leap motion,” 2017 14th IEEE Annual Consumer Communications & Networking Conference (CCNC), pp. 137-141, 2017.
[19]M. A. Seif, R. Umeda and H. Higa, “An attempt to control a 3D object in medical training system using leap motion,” 2017 International Conference on Intelligent Informatics and Biomedical Sciences (ICIIBMS), pp. 159-162, 2017.
[20]M. Alimanova, S. Borambayeva, D. Kozhamzharova, N. Kurmangaiyeva, D. Ospanova, G. Tyulepberdinova, G, Gaziz and A. Kassenkhan, “Gamification of hand rehabilitation process using virtual reality tools: Using Leap Motion for hand rehabilitation,” 2017 First IEEE International Conference on Robotic Computing (IRC), pp. 336-339, 2017.
[21]A. Begalinova and A. Shintemirov, “Design of embedded gesture recognition system for robotic applications,” 2014 IEEE 8th International Conference on Application of Information and Communication Technologies (AICT), pp. 1-4, 2014.
[22]C. Kertész, “Physiotherapy exercises recognition based on RGB-D human skeleton models,” 2013 European Modelling Symposium, pp. 21-29, 2013.
[23]A. S. Kundu, O. Mazumder, A. Dhar and S. Bhaumik, “Occupancy grid map generation using 360?? scanning xtion pro live for indoor mobile robot navigation,” 2016 IEEE First International Conference on Control, Measurement and Instrumentation (CMI), pp. 464-468, 2016.
[24]B. Çubukçu and U. Yüzgeç, “A physiotherapy application with MS kinect for patients with shoulder joint, muscle and tendon damage,” 2017 9th International Conference on Computational Intelligence and Communication Networks (CICN), pp. 225-228, 2017.
[25]Y. Gao, S. Esquivel, R. Koch, M. Ziegler, F. Zilly and J. Keinert, “A novel kinect V2 registration method for large-displacement environments using camera and scene constraints,” 2017 IEEE International Conference on Image Processing (ICIP), pp. 997-1001, 2017.
[26]S. M. Ayaz, D. Khan and M. Y. Kim, “3D handheld scanning based on multiview 3D registration using Kinect Sensing device,” 2017 IEEE International Conference on Multisensor Fusion and Integration for Intelligent Systems (MFI), pp. 330-335, 2017.
[27]J. Li, L. Qi, A. Zhao, X. Chen and J. Dong, “Dynamic long short-term memory network for skeleton-based gait recognition,” Symposia and Workshops on Ubiquitous, Autonomic and Trusted Computing (UIC-ATC), pp. 1-6, 2017.
[28]N. Anbarsanti, A. S. Prihatmanto, “Dance learning and recognition system based on hidden Markov model. a case study : aceh traditional dance,” 2014 IEEE 4th International Conference on System Engineering and Technology (ICSET), pp. 86-92, 2014.
[29]H. Xu, Y. Lee and C. Lee, “Activity recognition using Eigen-joints based on HMM,” 2015 12th International Conference on Ubiquitous Robots and Ambient Intelligence (URAI), pp. 300-305, 2015.
[30]P. T. Hai and H. H. Kha, “An efficient star skeleton extraction for human action recognition using hidden Markov models,” 2016 IEEE Sixth International Conference on Communications and Electronics (ICCE), pp. 351-356, 2016.
[31]A. Nava, L. Garrido and R. F. Brena, “Recognizing activities using a Kinect skeleton tracking and hidden markov models,” 2014 13th Mexican International Conference on Artificial Intelligence, pp. 82-88, 2014.
[32]C. Conly, A. Dillhoff and V. Athitsos, “Leveraging intra-class variations to improve large vocabulary gesture recognition,” 2016 23rd International Conference on Pattern Recognition (ICPR), pp. 907-912, 2016.
[33]J. Wu, J. Cheng and W. Feng, “3D dynamic gesture recognition based on improved HMMs with entropy,” 2014 IEEE International Conference on Information and Automation (ICIA), pp. 213-218, 2014.
[34]H. Hiyadi, F. Ababsa, C. Montagne, E. H. Bouyakhf and F. Regragui, “A depth-based approach for 3D dynamic gesture recognition,” 2015 12th International Conference on Informatics in Control, Automation and Robotics (ICINCO), pp. 103-110, 2015.
[35]Y. Gu, H. Do, Y. Ou and W. Sheng, “Human gesture recognition through a Kinect sensor,” 2012 IEEE International Conference on Robotics and Biomimetics (ROBIO), pp. 1379-1384, 2012.
[36]P. T. Bao, N. T. Binh and T. D. Khoa, “A new approach to hand tracking and gesture recognition by a new feature type and HMM,” 2009 Sixth International Conference on Fuzzy Systems and Knowledge Discovery, pp. 3-6, 2009.
[37]L. Hui and S. Yu-jie, “Research on face recognition algorithm based on improved convolution neural network,” 2018 13th IEEE Conference on Industrial Electronics and Applications (ICIEA), pp. 2802-2805, 2018.
[38]M. Azimi and A. Pacut, “The effect of gender-specific facial expressions on face recognition system's reliability,” 2018 IEEE International Conference on Automation, Quality and Testing, Robotics (AQTR), pp. 1-4, 2018.
[39]J. Shi, Y. Ming, C. Fan and L. Tian, “Face recognition algorithm based on multi-scale CLBP,” Symposia and Workshops on Ubiquitous, Autonomic and Trusted Computing (UIC-ATC), pp. 1-6, 2017.
[40]C. I. Orozco, F. Iglesias, M. E. Buemi and J. J. Berlles, “Real-time gender recognition from face images using deep convolutional neural network,” 7th Latin American Conference on Networked and Electronic Media (LACNEM 2017), pp. 7-11, 2017.
[41]C. S. Tan, C. K. Yong and Y. H. Tay, “Modeling financial ratios of Malaysian plantation stocks using Bayesian Networks,” 2012 IEEE Conference on Sustainable Utilization and Development in Engineering and Technology (STUDENT), pp. 7-12, 2012.
[42]N. Ramanan, S. Yang, S. Grannis and S. Natarajan, “Discriminative boosted Bayes networks for learning multiple cardiovascular procedures,” 2017 IEEE International Conference on Bioinformatics and Biomedicine (BIBM), pp. 870-873, 2017.
[43]X. Zhao, D. Huang, E. Dellandrea and L. Chen, “Automatic 3D facial expression recognition based on a bayesian belief Net and a statistical facial feature model,” 2010 20th International Conference on Pattern Recognition, pp. 3724-3727, 2010.
[44]M. Das and S. K. Ghosh, “A probabilistic approach for weather forecast using spatio-temporal inter-relationships among climate variables,” 2014 9th International Conference on Industrial and Information Systems (ICIIS), pp. 1-6, 2014.
[45]G. Ramos, A. Torres and M. Rios, “Analysis of electrical industrial systems using probabilistic networks,” IEEE Latin America Transactions, pp. 505-511, 2010.
[46]張鳳玲，「發展貝氏網路於基因網路的建構與探討」，慈濟大學醫學資訊研究所碩士論文，2005。
[47]J. Zhao, W. Wang, K. Sun and Y. Liu, “A bayesian networks structure learning and reasoning-based byproduct gas scheduling in steel industry,” IEEE Transactions on Automation Science and Engineering, pp. 1149-1154, 2014.
[48]Y. Ito, H. Izumi and C. Srinivasan, “Simultaneous learning of several Bayesian and Mahalanobis discriminant functions by a neural network with additional nodes,” The 2011 International Joint Conference on Neural Networks, pp. 733-740, 2011.
[49]S. Yang and K. C. Chang, “Comparison of score metrics for Bayesian network learning,” IEEE Transactions on Systems, Man, and Cybernetics - Part A: Systems and Humans, pp. 419-428, 2002.
[50]W. Lam, “Bayesian network refinement via machine learning approach,” IEEE Transactions on Pattern Analysis and Machine Intelligence, pp. 240-251, 1998.
[51]R. Chang and W. Wang, “Novel algorithm for Bayesian network parameter learning with informative prior constraints,” The 2010 International Joint Conference on Neural Networks (IJCNN), pp. 1-8, 2010.
[52]C. Echegoyen, J. A. Lozano, R. Santana and P. Larranaga, “Exact Bayesian network learning in estimation of distribution algorithms,” 2007 IEEE Congress on Evolutionary Computation, pp. 1051-1058, 2007.
[53]Y. You, J, Li and N. Xu, “A constrained parameter evolutionary learning algorithm for Bayesian network under incomplete and small data,” 2017 36th Chinese Control Conference (CCC), pp. 3044-3051, 2017.
[54]G. Xiao-guang, Y. Yu, G. Zhi-gao and C. Da-qing, “Bayesian approach to learn Bayesian networks using data and constraints,” 2016 23rd International Conference on Pattern Recognition (ICPR), pp. 3667-3672,2016.
[55]J. Shotton, A. Fitzgibbon, M. Cook, T. Sharp, M. Finocchio, R. Moore, A. Kipman and A. Blake, “Real-time human pose recognition in parts from single depth images,” IEEE Computer Vision and Pattern Recognition (CVPR) , pp. 1297-1304, 2011.
[56]https://www.microsoft.com/zh-tw/
[57]R. E. Walpole, R. H. Myers, S. L. Myers and K. E. Ye, “Probability and statistics for engineers and scientists (9th Edition),” Pearson Publisher, 2010.
[58]鄭祐昇，「運用資料探勘技術於知識圖之建立」，東海大學工業工程與經營資訊系碩士論文，2004。
[59]陳柏杉，「結合簡單貝氏網路與動態貝氏網路之效率分類方法」，國立東華大學資訊工程系碩士論文，2011。
[60]G. Casella and E. I. George, “Explaining the gibbs sampler,” The American Statistician, pp. 167-174, 1992.
[61]張育瑞，「一種人體姿勢命令辨識及其身分識別的強化式方法之研究」，國立虎尾科技大學電機工程學系碩士論文，2016。
[62]蘇俊麟，「基於影像之深度資訊的手勢辨識方法研究」， 國立虎尾科技大學電機工程學系碩士論文，2016。
[63]D. E. King, “Dlib-ml: A machine learning toolkit,” Journal of Machine Learning Research 10, pp. 1755-1758, 2009.
[64]http://8.share.photo.xuite.net/hwlib2007/18ca833/10321807/473762724_m.jpg
[65]https://www.cs.waikato.ac.nz/ml/weka/