指導教授：連豊力臺灣大學：電機工程學研究所張凱翔Chang, Kai-HsiangKai-HsiangChang2014-11-282018-07-062014-11-282018-07-062013http://ntur.lib.ntu.edu.tw//handle/246246/262898在視覺自動交通監控系統裡，需要一個已校正相機來計算追蹤車輛之速度及車長分類。許多基於消失點特性之校正演算法已提出過，像是On-Line、VWL、VVL與VVW，然而在某些情況下這些演算法會有各別不同的限制。這篇論文提出了包含調整函數之基於消失點特性相機校正演算法來克服先前方法之限制。 從視覺影像有效地量測三維資訊裡，多個相機同時校正是一個重要的步驟，每台相機之相對世界座標系統會被建立出來。根據每台相機之相對世界座標系統及校正結果，不同相機之視覺資訊可以被同時拿來使用。這篇論文提出了一個從兩台相機找出車輛之結果且驗證其結果之方法。 1) 基於消失點特性之相機校正演算法 提出之校正方法包含三個主要步驟，首先找出特徵點之座標，然後使用特徵點估計出消失點之座標，最後使用調整函數來找出最佳化之相機校正參數。 2) 從雙台相機建構車輛幾何特徵 提出之方法包含三個主要步驟，首先對從二號相機之影像資訊做影像處理統計之方法將車輛幾何特徵給建構出，然後從一號相機之影像用相機參數及影像處理方法找出車輛位置，最後根據一號相機及二號相機之關係來驗證所建構之車輛幾何特徵建構是否合理。A calibrated camera is required for computing the speeds and length-based classifications of tracked vehicles in vision-based automatic traffic-monitoring systems. Numerous vanishing-point-based (VPB) calibration methods are presented, such as On-Line, VWL, VVL, and VVW. However the previous methods all have their restrictions in some situations. This thesis proposes a VPB method including tuning function to overcome restrictions of previous methods. Camera calibration of multiple cameras is a critical step for conducting 3-D measurement effectively from visual images. Then the relative world coordinate system of each camera is established. Based on the relative world coordinate system and calibration of each camera, the visual information of image from different cameras can be used simultaneously. This thesis proposes a method to estimate the position of the vehicle which is taken by two different cameras, and verify the result. 1) A Vanishing-Point-Based Camera Calibration Method The proposed calibration method includes mainly three steps. In the first step, identify the coordinates of the feature points. Then estimate coordinates of the vanishing point the feature points. Finally, optimize the camera calibration parameters by using tuning function. Therefore the optimal camera calibration parameters are estimated. 2) Constructing Vehicular Geometric Features (VGF) from Two Cameras The proposed method includes mainly three steps. In the first step, construct VGF of the vehicle by image processing statistic method using image information from camera 2. In the second step, construct the position of the vehicle by calibration parameters and image processing from camera 1. Finally, verify the results of VGF by the relation of camera 1 and camera 2 and their calibration parameters. Therefore the VGF is constructed and the results are capable to analyze rational or not.摘要 i ABSTRACT iii Contents v List of Figures vii List of Tables xiii Chapter 1 Introduction 1 1.1 Motivation 1 1.2 Problem Formulation 3 1.3 Contribution of the Thesis 5 1.4 Organization of the Thesis 7 Chapter 2 Literature Survey 9 2.1 Vanishing-Point-Based Camera Calibration 9 2.2 Vehicle Detection by Using Shape Model 11 Chapter 3 Preliminary Models and Algorithms 13 3.1 Setup of the Camera 13 3.2 Pin-Hole Camera Model 16 3.3 Inverse Perspective Mapping 18 3.4 Edge Detectors 20 3.4.1 Sobel Edge Detector 21 3.4.2 Canny Edge Detector 22 Chapter 4 Vanishing-Point-Based Camera Calibration with Tuning Function 25 4.1 The Definition of Feature Points 27 4.2 Determination of Camera Calibration Parameters Groups with Vanishing Point 29 4.2.1 Vanishing Point Estimation 30 4.2.2 Projective Geometry 32 4.3 Optimizing Results of Camera Calibration by Using Tuning Function 35 Chapter 5 Constructing Vehicular Geometric Features (VGF) from Two Cameras 41 5.1 Two Cameras System Architecture and Camera Calibration 45 5.2 Constructing VGF from Camera 2 47 5.2.1 Vehicular Edge Extraction with Background Edge Removal 50 5.2.2 Construction of VGF from Camera 2 52 5.3 Constructing VGF from Camera 1 72 5.3.1 Partial Vehicular Edge Extraction with Background Edge Removal and Capture Filter 73 5.3.2 Construction of VGF from Camera 1 75 5.4 Verifying Results of VGF Construction from Two Cameras 81 5.4.1 Feature Vector 82 5.4.2 Verification of VGF Construction Results from Two Cameras Ideally 84 5.4.3 Verification of VGF Construction Results from Two Cameras in Reality 88 Chapter 6 Simulations and Experiments 91 6.1 Vanishing-Point-Based Camera Calibration with Tuning Function 91 6.2 Constructing VGF from Two Cameras 101 6.2.1 Constructing VGF from Camera 2 103 6.2.2 Constructing VGF from Camera 1 140 6.2.3 Verifying Results of VGF Construction from Two Cameras 151 Chapter 7 Conclusions and Future Works 155 7.1 Conclusions 155 7.2 Future Works 157 References 1594054780 bytesapplication/pdf論文公開時間：2018/11/05論文使用權限：同意有償授權(權利金給回饋學校)相機校正單眼相機多個相機基於視覺感測3D量測交通監控消失點車輛偵測thesishttp://ntur.lib.ntu.edu.tw/bitstream/246246/262898/1/ntu-102-R00921065-1.pdf