傅立成臺灣大學:資訊工程學研究所陳意函Chen, Yi-HanYi-HanChen2010-05-172018-07-052010-05-172018-07-052009U0001-1408200911284600http://ntur.lib.ntu.edu.tw//handle/246246/183343在一個智慧家庭裡，理解使用者的喜好並提供相對應的服務是相當重要地。另一方面，從環境中的資訊辨識出使用者當下的行為是一項艱鉅的挑戰，為了提供使用者更合適的服務，了解使用者的行為將能更正確地提供使用者所需要的服務。在過去的研究裡，使用者的偏好以及其行為的辨識在智慧家庭中通常是分開被討論，在這篇論文裡，我們首次嘗試發展出一個整合性系統來建立出個人喜好模型和其行為模型之間的關聯性，透過了解使用者的偏好去提升行為辨識在動態環境下的準確度。更明確地說，也就是透過了解使用者當下的行為，我們可以藉由這項資訊來學習使用者的偏好，並根據學習好的模型來提供更準確地服務給使用者，另外，使用者對所提供服務的反應則可回饋給系統用來修正並調整已經學習好的模型，行為模型也同時透過分析使用者的反應來做調整。除此之外，我們進一步從單人的整合系統中設計出一個多人的整合系統，其成果皆呈現在實驗中。Understanding a user’s preferences and then providing corresponding services is substantial in a smart home environment nowadays. On the other hand, reliable recognition of activities from cluttered sensory data is challenging and important as well for a smart home to provide more desirable services. Traditionally, preference learning and activity recognition for a smart home system were dealt with separately. In this thesis, we aim to develop a hybrid system which is the first trial to model the relationship of a preference model and an activity model so that the causal relation among activities and personal preferences can assist to recover the accuracy of activity recognition in the dynamic environment. Specifically, on-going activity which a user performs in this work is regarded as high level contexts to assist in learning the user’s preference model, Based on the learned model, the smart home system provides services to the user so that the hybrid system can better interact with the user and also gain his/her feedback to adjust the learned preference model. Afterwards, both of the activity model and preference model will be simultaneously adapted by the analysis of the feedback. In addition, we further design a multi-user hybrid system, which is extended from single-user hybrid system, to deal with the interactions among users in a multi-user environment. The experimental results are provided to show the effectiveness of the proposed approach.口試委員會審定書 #謝 i文摘要 iiBSTRACT iiiONTENTS ivIST OF FIGURES viiIST OF TABLES ixhapter 1 Introduction 1.1 Motivation 1.2 Challenges 4.3 Related Work 4.4 Objective 6.5 Thesis Organization 6hapter 2 Preliminaries 8.1 System Overview 8.2 Dynamic Bayesian Network (DBN) 10.2.1 Representation 10.2.2 Inference 11.2.3 Learning 13.3 Coupled Hidden Markov Models (CHMMs) 15.3.1 Representation 15.3.2 Parameter Learning 16.3.3 Inference 17.4 Particle Filter 19hapter 3 Hybrid System of an Activity Model and a Preference Model 20.1 Overview 20.2 Environmental Sensors 21.3 Activity Model 24.3.1 Data Preprocessing and Feature Extraction 25.3.2 Feature Selection 26.3.3 Model Training 28.3.4 Activity Recognition 29.4 Preference Model 29.4.1 Context Interpret 30.4.2 Personal Preference Modeling 31.4.3 Inference 33.5 Hybrid System 34.5.1 Context of Activities 34.5.2 Simultaneous Activity Recognition and Service Providing 35.5.3 Factors of Dynamic Change 38.5.4 Adaptation of Preference Model 40.5.5 Preference Model assisted Activity Learning 41hapter 4 Hybrid System of the Multi-user’s Environment 44.1 Overview 44.2 Multi-target Tracking System 45.2.1 Bayes Filter Approach 46.2.2 Particle Filter Approach 47.2.3 Data Association 48.3 Multi-user Activity Model 49.4 Multi-user Preference Model 51.4.1 Representation 53.4.2 Parameters Learning 54.4.3 Inference 54.5 Multi-user Hybrid System 55hapter 5 System Evaluation 57.1 Experiment Environment 57.2 Experimental Result and Description 59.2.1 Single-User Environment 59.2.2 Multi-User Environment 61hapter 6 Conclusion 69.1 Summary 69.2 Future Work 70EFERENCE 71application/pdf4927702 bytesapplication/pdfen-US行為辨識動態貝氏網路偏好推論Activity RecognitionDynamic Bayesian NetworkPreference Inferencethesishttp://ntur.lib.ntu.edu.tw/bitstream/246246/183343/1/ntu-98-R96922064-1.pdf