Embedded Computer Vision (eBook)

Foreword 6
Preface 8
Embedded Computer Vision 8
Target Audience 9
Organization of the Book 10
Overview of Chapters 10
How This Book Came About 12
Outlook 13
Acknowledgements 14
Contents 15
List of Contributors 22
Introduction 26
Hardware Considerations for Embedded Vision Systems 27
1.1 The Real-Time Computer Vision Pipeline 27
1.2 Sensors 29
1.3 Interconnects to Sensors 33
1.4 Image Operations 35
1.5 Hardware Components 36
1.6 Processing Board Organization 46
1.7 Conclusions 48
References 49
Design Methodology for Embedded Computer Vision Systems 51
2.1 Introduction 51
2.2 Algorithms 54
2.3 Architectures 55
2.4 Interfaces 57
2.5 Design Methodology 59
2.6 Conclusions 67
References 67
We Can Watch It for You Wholesale 72
3.1 Introduction to Embedded Video Analytics 72
3.2 Video Analytics Goes Down-Market 74
3.3 How Does Video AnalyticsWork? 79
3.4 An Embedded Video Analytics System: by the Numbers 89
3.5 Future Directions for Embedded Video Analytics 93
3.6 Conclusion 97
References 98
Advances in Embedded Computer Vision 100
Using Robust Local Features on DSP-Based Embedded Systems 101
4.1 Introduction 101
4.2 RelatedWork 103
4.3 Algorithm Selection 104
4.4 Experiments 109
4.5 Conclusion 119
References 121
Benchmarks of Low-Level Vision Algorithms for DSP, FPGA, and Mobile PC Processors 123
5.1 Introduction 123
5.2 RelatedWork 125
5.3 Benchmark Metrics 125
5.4 Implementation 126
5.5 Results 139
5.6 Conclusions 140
References 141
SAD-Based Stereo Matching Using FPGAs 143
6.1 Introduction 143
6.2 RelatedWork 144
6.3 Stereo Vision Algorithm 145
6.4 Hardware Implementation 147
6.5 Experimental Evaluation 151
6.6 Conclusions 159
References 159
Motion History Histograms for Human Action Recognition 161
7.1 Introduction 161
7.2 RelatedWork 163
7.3 SVM-Based Human Action Recognition System 164
7.4 Motion Features 165
7.5 Dimension Reduction and Feature Combination 170
7.6 System Evaluation 172
7.7 FPGA Implementation on Videoware 178
7.8 Conclusions 182
References 183
Embedded Real-Time Surveillance Using Multimodal Mean Background Modeling 185
8.1 Introduction 185
8.2 RelatedWork 186
8.3 Multimodal Mean Background Technique 188
8.4 Experiment 190
8.5 Results and Evaluation 192
8.6 Conclusion 196
References 197
Implementation Considerations for Automotive Vision Systems on a Fixed- Point DSP 198
9.1 Introduction 198
9.2 Fixed-Point Arithmetic 203
9.3 Process of Dynamic Range Estimation 203
9.4 Implementation Considerations for Single-Camera Steering Assistance Systems on a Fixed- Point DSP 207
9.5 Results 211
9.6 Conclusions 214
References 215
Towards OpenVL: Improving Real-Time Performance of Computer Vision Applications 216
10.1 Introduction 216
10.2 RelatedWork 218
10.3 A Novel Software Architecture for OpenVL 222
10.4 Example Application Designs 232
10.5 Conclusion and Future Work 235
10.6 Acknowledgements 236
References 236
Looking Ahead 238
Mobile Challenges for Embedded Computer Vision 239
11.1 Introduction 239
11.2 In Search of the Killer Applications 241
11.3 Technology Constraints 244
11.4 Intangible Obstacles 250
11.5 Future Direction 252
References 253
Challenges in Video Analytics 256
12.1 Introduction 256
12.2 Current Technology and Applications 257
12.3 Building Blocks 263
12.4 Embedded Implementations 267
12.5 Future Applications and Challenges 269
12.6 Summary 273
References 274
Challenges of Embedded Computer Vision in Automotive Safety Systems 276
13.1 Computer Vision in Automotive Safety Applications 276
13.2 Literature Review 277
13.3 Vehicle Cueing 278
13.4 Feature Extraction 287
13.5 Feature Selection and Classification 293
13.6 Experiments 295
13.7 Conclusion 297
References 297
Index 299