黃漢邦臺灣大學：機械工程學研究所林俊廷Lin, Chun-TingChun-TingLin2007-11-282018-06-282007-11-282018-06-282005http://ntur.lib.ntu.edu.tw//handle/246246/61453本文的主要目的為發展多物件影像追蹤與多角度人臉偵測與辨識系統。我們提出Multi-CAMSHIFT來實現多物件追蹤，利用所感興趣的機率分布特性，例如：顏色、形狀，快速追蹤出物件輪廓以作為候選區域。整個系統架構於多解析度機制之上，可以有效改善系統效能並且降低龐大的運算量。配合多組主成分分析(PCA)和支持向量機(SVM)，利用不同角度的奇異臉分析，結合成多角度人臉偵測與辨識模組，對於不同人臉姿勢加以分類與身分辨識。 我們的系統可應用於複雜背景以及即時追蹤，並且利用機率模型的更新機制，有效解決緩慢光源的變化。我們將上述的方法以及理論，成功地實現多角度人臉追蹤與辨識，並且應用於監測系統、人形追蹤和人臉辨識門禁等各系統。This thesis, aims to develop a system for multiple objects tracking and multi-view faces detection and recognition. We propose a novel method (Multi-CAMSHIFT), which is based on the characteristics of color and shape probability distribution, to solve the tracking problems for multiple objects. The tracker is used to get the candidate regions by outlining the interested probability distribution. The system performance is further improved by using multi-resolution framework and computation reduction. The principal component analysis (PCA) and support vector machine (SVM) are integrated to form the multi-view faces detection and recognition module for classifying different face poses and identities. Beside color information, the gray background image is used to locate the human head in the region of tracking pedestrian based on probability distribution rule. The rule can also be used for skin color face tracking to remove background region (non-face region). Since the proposed Multi-CAMSHIFT (MCAMSHIFT) is computationally efficient, it can work in complex background and track in real-time. The slowly changing lighting condition is effectively resolved using probability model update. From experiments, the proposed MCAMSHIFT was successfully applied to multi-view faces tracking and recognition. It can also be applied to surveillance system, pedestrian tracking and face guard systems.摘要 i Abstract ii List of Tables vi List of Figures vii Chapter 1 Introduction 1 1.1. Motivation 1 1.2. Related Works 2 1.2.1. Object Tracking 2 1.2.2. Face Tracking and Recognition 4 1.3. Objectives and Contributions 6 1.4. Thesis Organization 9 Chapter 2 Background Knowledge 10 2.1. Color Space Used for Skin Modeling 10 2.2. The CAMSHIFT Algorithm 12 2.2.1. Introduction to the CAMSHIFT Algorithm 12 2.2.2. Mass Center Calculation 13 2.2.3. Probability Distribution 15 2.3. Principal Components Analysis (PCA) 16 2.4. Support Vector Machine (SVM) 20 2.4.1. Structural Risk Minimization 20 2.4.2. Introduction to SVMs 21 Chapter 3 Multiple Objects Tracking 27 3.1. Interested probability Modeling 27 3.1.1. Skin Color Probability Modeling 27 3.1.2. Background Probability Modeling 31 3.2. Probability Model Update 33 3.2.1. Adaptive Skin Color Probability Model Update 34 3.2.2. Adaptive Background Probability Model Update 34 3.3. Modified CAMSHIFT Algorithm 36 3.3.1. Interested probability Enhancement 38 3.3.2. Multi-Resolution Framework 40 3.3.3. Initial Block Searching in Small Resolution 42 3.3.4. Search Window of CAMSHIFT 44 3.3.5. Center Tendency 45 3.4. Multi-CAMSHIFT Algorithm (MCAMSHIFT) 47 3.4.1. Sort Indexes of MCAMSHIFT 50 Chapter 4 Multi-View Faces Detection and Recognition 52 4.1. Face Pattern Enhancement and Classification 52 4.1.1. Pattern Histogram Equalization Enhancement 53 4.1.2. Eigenfaces 55 4.1.3. Mask Filter 56 4.1.4. SVM Data Scaling 59 4.1.5. PCA and SVMs Face Classifier 60 4.2. Multi-View Faces Module 64 4.2.1. Multi-View Faces Representation 65 4.2.2. Combined PCA-SVMs Module 66 4.3. Multiple Faces Tracking and Recognition 70 4.3.1. PID Control Theorem in Pan-Tilt System 72 4.3.2. Moving Pedestrians Tracking and Heads Tracking 73 Chapter 5 Applications and Experimental Results 75 5.1. System Overview 75 5.2. Applications 77 5.2.1. Face Tracking and Recognition 79 5.2.2. Surveillance System and Guard System 81 5.3. Performance of MCAMSHIFT Tracking 84 5.4. Face Recognition Experiments 88 5.4.1. Static Multi-View Faces Recognition Experiments 89 5.4.2. Dynamic Multi-View Faces Recognition Experiments 90 Chapter 6 Conclusions 93 6.1. Conclusions 93 6.2. Future Works 94 References 953960812 bytesapplication/pdfen-US人臉追蹤偵測辨識多角度人臉主成分分析支持向量機facetrackingdetectionrecognitionmulti-viewPCASVMthesishttp://ntur.lib.ntu.edu.tw/bitstream/246246/61453/1/ntu-94-R92522819-1.pdf