指導教授：陳永耀臺灣大學：電機工程學研究所王遵羲Wang, Thun-HsiThun-HsiWang2014-11-282018-07-062014-11-282018-07-062014http://ntur.lib.ntu.edu.tw//handle/246246/262905近年來，社會型態逐漸走向老齡化與少子化，越來越多的老人需要在家中受到照護。 然而，在照護人力日漸不足的情況下，自動化的居家看護系統已逐漸受到重視。考慮到居家看護系統在人們心中的接受度與實用性，該系統必需在不干擾使用者日常生活的情況下運行。此外，針對老年人在家中所面臨的問題，我們主要專注在睡眠行為的監測與跌倒行為的辨識上。因此，本論文提出一套由單一深度攝影機所建構的睡眠監侧與跌倒偵測系統。為了達成準確的空間關係推算，我們使用深度攝影機所提供的深度影像與視角參數來轉換出世界座標系，並由深度影像與動態量分析來擷取景像中的主體。然而藉由推算床鋪區間與目標的空間相關係數後，可明確的判定目標物的上下床時機，而深度影像也可隨時紀錄目標物在床上的動作變化以利於做細部分析。同時我們也使用目標物的實際高度變化與在影像上的歷史變化來結合條件式判定目標物有無發生跌倒的情況。最後，上下床偵測與睡眠動作量偵測的成功率達到98.4%，跌倒偵測的準確度達到96.8%。Along with the population ageing and birth bust, more and more elderly people need to be cared in their daily living. Considering the acceptance rate and practicability for elderly people, home care systems should work under the conditions that people should not be disturbed by these systems in their daily living. In particular, we focus on the research of sleep behavior and fall event when elderly people are individual at home. Therefore, we propose a sleep monitoring and fall detection system using single depth camera. To calculate correct spatial relationship between human and furniture in an image, a coordinate transformation approach based on depth information and visual angles is used to build up the world coordinate system. By calculating the spatial relationship between human and bed region, people’s sleep behavior and movement on the bed can be recognized precisely. In fall detection, detection methods are mainly composed of motion analysis and posture height. The results of proposed system are 98.4% accuracy rate in in-bed detection and sleep motion detection, and the overall accuracy of fall detection is 96.8%.誌謝 i 中文摘要 ii ABSTRACT iii CONTENTS iv LIST OF FIGURES vii LIST OF TABLES xiv Chapter 1 Introduction 1 1.1 Motivation 2 1.2 Previous Work of Home Care System 2 1.3 Problem Definition 5 1.4 Proposed Approach 6 1.5 Thesis Overview 10 Chapter 2 Study of Sleep Monitoring and Fall Detection 11 2.1 Sleep Monitoring 11 2.1.1 Invasive Sensors 13 2.1.2 Contact Sensors 15 2.1.3 Vision-based System 18 2.2 Fall Detection 22 2.2.1 Audio-based Approach 23 2.2.2 2D Vision-based Approach 25 2.2.3 3D Vision-based Approach 31 2.3 Comparisons and Summary 37 Chapter 3 Foreground Extraction and Motion Analysis 39 3.1 IR Sensor and Camera Calibration 40 3.2 Foreground Extraction 44 3.3 Background Revision 47 3.4 Features Extraction 50 3.5 Motion Trajectory Analysis 53 Chapter 4 Behavior Monitoring in Bedroom Environment 55 4.1 System Overview 55 4.2 Sleep Monitoring 56 4.2.1 Preprocessing for Setting Bed Space 57 4.2.2 In-bed Detection 60 4.2.3 Movement Detection 63 4.2.4 Awaking Warning 65 4.2.5 A Concept of People Counting 68 4.3 Fall Detection 72 4.3.1 Height Variation and Normalized Value of MHI 72 4.3.2 Rules for Fall Detection 74 Chapter 5 Experimental Results 78 5.1 Environment Setting 79 5.2 Result of Sleep Monitoring 80 5.2.1 In-bed Detection 81 5.2.2 Motion Detection 83 5.2.3 Awaking Warning 85 5.3 Results of Fall Detection 86 5.4 Comparisons and Discussions 90 Chapter 6 Conclusions and Future Work 93 REFERENCES 946072136 bytesapplication/pdf論文公開時間：2019/08/21論文使用權限：同意無償授權電腦視覺深度攝影機居家看護睡眠監測跌倒偵測thesishttp://ntur.lib.ntu.edu.tw/bitstream/246246/262905/1/ntu-103-R01921090-1.pdf