Computer Vision for Driver Assistance (eBook)

Mahdi Rezaei is Assistant Professor at Qazvin Islamic Azad University, Iran, and Honorary Academic Staff at the University of Auckland, New Zealand. He has a PhD in Computer Science and was awarded the Best Thesis Award from the University of Auckland. His research interests include computer vision, pattern recognition, and advanced driver assistance systems. Rezaei is the author of numerous contributions to top publications, including IEEE Transactions on Intelligent Transportation Systems and IEEE Conference on Computer Vision and Pattern Recognition, CVPR. Reinhard Klette, Fellow of the Royal Society of New Zealand, is Professor at the Auckland University of Technology, New Zealand. He previously held positions at the University of Auckland, the Technical University of Berlin, and the Academy of Sciences Berlin. His research interests include computer vision, pattern recognition, and algorithm design. From 2003 to 2008, he was Associate Editor of IEEE Transactions on Pattern Analysis and Machine Intelligence.

Preface 7
Contents 10
Symbols 13
1 Vision-Based Driver-Assistance Systems 15
1.1 Driver-Assistance Towards Autonomous Driving 15
1.2 Sensors 16
1.3 Vision-Based Driver Assistance 18
1.4 Safety and Comfort Functionalities 21
1.5 VB-DAS Examples 22
1.6 Current Developments 26
1.7 Scope of the Book 29
2 Driver-Environment Understanding 33
2.1 Driver and Environment 33
2.2 Driver Monitoring 34
2.3 Basic Environment Monitoring 39
2.4 Midlevel Environment Perception 44
3 Computer Vision Basics 51
3.1 Image Notations 51
3.2 The Integral Image 53
3.3 RGB to HSV Conversion 54
3.4 Line Detection by Hough Transform 55
3.5 Cameras 57
3.6 Stereo Vision and Energy Optimization 58
3.7 Stereo Matching 61
4 Object Detection, Classification, and Tracking 64
4.1 Object Detection and Classification 64
4.2 Supervised Classification Techniques 66
4.2.1 The Support Vector Machine 66
4.2.2 The Histogram of Oriented Gradients 72
4.2.3 Haar-Like Features 76
4.3 Unsupervised Classification Techniques 81
4.3.1 k-Means Clustering 81
4.3.2 Gaussian Mixture Models 85
4.4 Object Tracking 91
4.4.1 Mean Shift 93
4.4.1.1 Mean Shift Tracking 95
4.4.2 Continuously Adaptive Mean Shift 97
4.4.3 The Kanade–Lucas–Tomasi (KLT) Tracker 98
4.4.4 Kalman Filter 102
4.4.4.1 Filter Implementation 104
4.4.4.2 Tracking by Prediction and Refinement 106
5 Driver Drowsiness Detection 108
5.1 Introduction 108
5.2 Training Phase: The Dataset 110
5.3 Boosting Parameters 112
5.4 Application Phase: Brief Ideas 112
5.5 Adaptive Classifier 115
5.5.1 Failures Under Challenging Lighting Conditions 115
5.5.2 Hybrid Intensity Averaging 117
5.5.3 Parameter Adaptation 118
5.6 Tracking and Search Minimization 120
5.6.1 Tracking Considerations 120
5.6.2 Filter Modelling and Implementation 121
5.7 Phase-Preserving Denoising 123
5.8 Global Haar-Like Features 124
5.8.1 Global Features vs. Local Features 125
5.8.2 Dynamic Global Haar Features 127
5.9 Boosting Cascades with Local and Global Features 127
5.10 Experimental Results 128
5.11 Concluding Remarks 138
6 Driver Inattention Detection 140
6.1 Introduction 140
6.2 Asymmetric Appearance Models 142
6.2.1 Model Implementation 142
6.2.2 Asymmetric AAM 144
6.3 Driver's Head-Pose and Gaze Estimation 146
6.3.1 Optimized 2D to 3D Pose Modelling 147
6.3.2 Face Registration by Fermat-Transform 149
6.4 Experimental Results 152
6.4.1 Pose Estimation 152
6.4.2 Yawning Detection and Head Nodding 152
6.5 Concluding Remarks 157
7 Vehicle Detection and Distance Estimation 159
7.1 Introduction 159
7.2 Overview of Methodology 161
7.3 Adaptive Global Haar Classifier 164
7.4 Line and Corner Features 167
7.4.1 Horizontal Edges 168
7.4.2 Feature-Point Detection 169
7.5 Detection Based on Taillights 171
7.5.1 Taillight Specifications: Discussion 171
7.5.2 Colour Spectrum Analysis 173
7.5.3 Taillight Segmentation 174
7.5.4 Taillight Pairing by Template Matching 175
7.5.5 Taillight Pairing by Virtual Symmetry Detection 177
7.6 Data Fusion and Temporal Information 180
7.7 Inter-vehicle Distance Estimation 183
7.8 Experimental Results 186
7.8.1 Evaluations of Distance Estimation 187
7.8.2 Evaluations of the Proposed Vehicle Detection 188
7.9 Concluding Remarks 199
8 Fuzzy Fusion for Collision Avoidance 201
8.1 Introduction 201
8.2 System Components 203
8.3 Fuzzifier and Membership Functions 204
8.4 Fuzzy Inference and Fusion Engine 207
8.4.1 Rule of Implication 208
8.4.2 Rule of Aggregation 208
8.5 Defuzzification 209
8.6 Experimental Results 209
8.7 Concluding Remarks 216
Erratum to: Computer Vision for Driver Assistance: Simultaneous Traffic and Driver Monitoring 218
Bibliography 219
Index 233