陳永耀臺灣大學：電機工程學研究所李冠德Li, Kun-TeKun-TeLi2007-11-262018-07-062007-11-262018-07-062005http://ntur.lib.ntu.edu.tw//handle/246246/53317辨識人體運動行為是現今電腦視覺中一項有趣的研究主題，已被廣泛地運用在虛擬實境、監控系統、人機介面、人體動作分析與多媒體壓縮等應用上。在含人體運動的連續畫面中，辨識人體運動行為主要可分兩大部份:人體資訊特徵的取得及人體動作行為的解析表達。人體資訊特徵取得的困難在於如何從影像中找出人體部份，而人體動作行為解析表達的挑戰在於如何利用人體資訊特徵辨識多變複雜的人體動作。 本篇論文針對人體動作行為解析表達這方面，提出一套針對居家看護的人體動作辨識系統。其做法是依據人類辨識動作的直觀判斷，設定每個動作之規則條件，再藉由判斷人體資訊特徵是否符合動作條件而辨識之，且為了居家看護的需求，特別要求跌倒等危險動作的判斷。本系統利用3D人體動畫軟體取代影像的人體資訊特徵取得，藉由模糊理論比較人體資訊與動作規則之間的符合程度，以符合程度最高的動作為系統之輸出。實驗中將利用3D人體動畫與程式輸出的比對，證明本方法可以成功的辨識解析人體動作。Recognition of human activities has become an interesting research topic in computer vision, with a wide range of applications in virtual reality, surveillance systems, human-computer interface, human motion analysis and multimedia compression, etc. Recognition of human activities mainly involves two aspects: acquirement of human motion information and representation of human activities. The difficulties of human information acquirement are to extract human body information and characteristics from the image sequence. The challenge of human activities representation is how to utilize human information to recognize complex and varied human actions. This thesis focuses on the representation of human activities. A human action recognition system for home-care is presented. The methodology of recognition system depends on the intuitional identification of human to design individual rules for each action. According to the correlation between rules and human information, actions can be recognized. This system extracts the human body information from the image sequence which is generated by a 3D human body animation software. Fuzzy set theory is utilized to derive the weighting of each action. By comparing the weighting of each action, the action with the largest weighting is chosen as the output of this recognition system. According to a series of experiments, the comparison between 3D human animation and program results shows that the proposed system can recognize human actions successfully.Abstract i 摘要 ii Contents iii List of Figures v List of Tables vii Chapter 1 Introduction 1 1.1 Motivation 1 1.2 Problem Definition 2 1.3 Previous Methods 3 1.4 Approach 5 1.5 Thesis Overview 7 Chapter 2 Human Activities Identification 9 2.1 Human Body Modeling 10 2.1.1 View-Based Approaches 11 2.1.2 Model-Based Approaches 12 2.2 Human Activities Recognition Approaches 16 2.2.1 Human Action Recognition Approaches 17 2.2.2 Behavior Recognition Schemes with Domain Knowledge 23 2.3 Summary 26 Chapter 3 Design of Human Action Recognition 29 3.1 Characteristics of Action Identification by Human 30 3.1.1 Recognition by Features of Actions 30 3.1.2 Multiple Features for One Action 31 3.1.3 Fuzziness in Action Identification 31 3.2 Classification of Human Actions 32 3.3 Fuzzy Set Theory 35 3.3.1 Introduction to Fuzzy Set Theory 36 3.3.2 Fuzzy Set 37 3.3.3 Membership Function 39 3.3.4 Fuzzy Rule 40 3.3.5 Fuzzy Inference 41 3.4 Fuzzy Action Recognition 42 3.4.1 Fuzzification 43 3.4.2 Rule Base 44 3.4.3 Fuzzy Inference Engine 45 3.5 Rule Design 48 3.6 Summary 50 Chapter 4 Implementation of Human Action Recognition System 51 4.1 System Architecture 51 4.2 Front-End Simulation 54 4.2.1 The Structure of Human Model 55 4.2.2 Biovision Hierarchical File 58 4.3 Recognition Factors Derivation 59 4.4 Summary 63 Chapter 5 Simulation and Results 65 5.1 Basic Test 66 5.2 Special Test for Home Care System 70 5.3 Composite Test 73 5.4 Summary 75 Chapter 6 Conclusion 77 References 791214318 bytesapplication/pdfen-US人體動作辨識模糊理論居家看護human action recognitionfuzzy set theoryhome-care systemthesishttp://ntur.lib.ntu.edu.tw/bitstream/246246/53317/1/ntu-94-R92921012-1.pdf