陳永耀臺灣大學：電機工程學研究所陳信名Chen, Hsin-MingHsin-MingChen2007-11-262018-07-062007-11-262018-07-062005http://ntur.lib.ntu.edu.tw//handle/246246/53018智慧型監測系統應用於許多場合，也是個有趣的研究主題，經由智慧型的判斷，可以降低許多人力成本以及人為疏失。 本論文提出以雙計數器為基礎之移動物體分割演算法,並希望應用於居家看護系統。此方法作為人體動作判別的前端處理。在移動物體分割部份，首先利用差異值統計圖去自動取得門檻值以適應於不同的拍攝環境，有效地將移動的物體從畫面中分離出來，並利用形態學的運算子去補償切割後的畫面和消除陰影的影響，最後，本論文利用統計的概念去更新背景的資訊，提出一個雙計數器背景註冊法，建立具有可靠性的背景，使系統能夠有效的應用在各種不同的監測環境之下。In this thesis, we designed a two-counter background registration method on moving object segmentation which is an important front-end process for human activity recognition. In moving object segmentation, we can use histograms of different values to determine the threshold value for automatic noise elimination in different picture frames. Moving objects are separated from the background in the current frame effectively with the proposed threshold method. The morphological operator is then used to eliminate the background noise of the segmentation results and shadows of the moving objects. Finally, we use the statistic conception to update the background information. We propose the Two-Counter background registration algorithm to reconstruct the reliable background information effectively for different surveillance environments.Abstract i 摘要 ii Contents iii List of Figures vii List of Tables xiii Chapter 1 Introduction 1 1.1. Motivation 1 1.2. Outline of Moving Object Segmentation 3 1.3. Thesis Overview 5 Chapter 2 Research Background and Relative Work 7 2.1. Segmentation Algorithm Review 7 2.1.1. Temporal Segmentation 8 2.1.2. Spatial Segmentation 8 2.1.3. Spatio-Temporal Segmentation 10 2.1.4. Motion Estimation Segmentation 11 2.1.5. Background Subtraction Segmentation 12 2.1.6. Comparison of Different Segmentation Algorithm 13 2.2. Previous Work of Moving Object Segmentation 14 2.2.1. Survey of Threshold Decision Algorithm 15 2.2.2. Survey of Shadow Effect Modeling 15 2.2.3. Survey of Background Updating Algorithm 18 Chapter 3 New Method of Moving Object Segmentation 21 3.1. System Architecture of Human Detection 22 3.2. Auto-Tuning Threshold Algorithm 23 3.2.1. Threshold Decision Problem 23 3.2.2. The Rules of Auto-Tuning Threshold 27 3.2.3. Experiment Results of Auto-Tuning Threshold 29 3.3. Shadow Cancellation 41 3.3.1. Shadow Cancellation Algorithm 41 3.3.2. Experiment Results of Shadow Cancellation 43 3.4. Morphological Process 46 3.4.1. The Morphological Operator 46 3.4.2. Boundary Process and Edge Detection 48 3.4.3. Experiment Result of Morphological Process 50 3.5. Two-Counter Background Registration Algorithm 51 3.5.1. Previous Work of Background Registration Algorithm 51 3.5.2. Statistic Approach 56 3.5.3. Two-Counter Background Registration Algorithm 60 3.5.4. Two-Counter Background Registration Algorithm with Upper Limit 68 3.5.5. Comparison Table of Different Background Registration Algorithm 69 Chapter 4 Experiment Results 71 4.1. The Performance of Two-Counter Background Registration 71 4.2. The Two-Counter Background Registration Algorithm with Background Index 75 4.3. The Registration Speed with Upper Limit and Without Upper Limit 79 Chapter 5 Conclusion 89 References 912816164 bytesapplication/pdfen-US門檻值決定背景註冊智慧型監測系統threshold determinesbackground registrationIntelligent surveillance systemthesishttp://ntur.lib.ntu.edu.tw/bitstream/246246/53018/1/ntu-94-R92921059-1.pdf