Person Re-Identification

Preface......Page 5
Acknowledgments......Page 8
Contents......Page 9
Contributors......Page 12
1 The Re-identification Challenge......Page 16
 1.1 Introduction......Page 17
 1.2 Re-identification Pipeline......Page 18
  1.2.1 A Taxonomy of Methods......Page 19
  1.3.2 Model and System Design......Page 20
  1.3.3 Data and Evaluation......Page 21
  1.4.1 On Feature Representation......Page 22
  1.4.2 On Model Learning......Page 26
  1.4.3 From Closed- to Open-World Re-identification......Page 28
  1.5.1 Multi-spectral and Multimodal Analysis......Page 31
 1.6 Further Reading......Page 32
 References......Page 33
Part IFeatures and Representations......Page 36
 2.1 Introduction......Page 37
 2.2 Related Work......Page 39
  2.3.1 Low-Level Biologically Inspired Features (BIF)......Page 40
  2.3.2 BiCov Descriptor......Page 42
  2.3.3 BiCoV Analysis......Page 44
  2.3.4 Experiments......Page 45
 2.4  Fisher Vector Encoded Local Descriptors for Person Re-identification......Page 47
  2.4.1 Local Image Descriptor......Page 48
  2.4.2 Extending the Descriptor......Page 49
  2.4.3 Experiments......Page 50
 References......Page 54
 3.1 Introduction......Page 57
 3.2 Related Work......Page 60
  3.3.1 Image Gathering and Selection......Page 62
  3.3.3 Symmetry-Based Silhouette Partition......Page 63
  3.3.4 Accumulation of Local Features......Page 66
 3.4 Signature Matching......Page 69
  3.4.1 Analysis......Page 70
 3.5 SDALF for Tracking......Page 71
  3.5.1 Particle Filter......Page 72
 3.6 Experiments......Page 73
  3.6.1 Results: Re-identification......Page 74
  3.6.2 Results: Tracking......Page 79
 References......Page 81
 4.1 Introduction......Page 84
 4.2 Related Work......Page 85
  4.3.1 Riemannian Geometry......Page 87
  4.3.2 Mean Riemannian Covariance......Page 90
  4.4.1 General Scheme for Appearance Extraction......Page 91
  4.4.2 MRCG Model......Page 92
  4.4.3 COSMATI Model......Page 94
  4.4.4 Appearance Matching......Page 97
  4.5.1 Experimental Setup......Page 98
  4.5.2 Results......Page 100
 References......Page 102
 5.1 Introduction......Page 105
  5.2.2 Attributes as Representation......Page 107
  5.2.3 Attributes for Identification......Page 108
  5.3.1 Ontology Selection......Page 109
  5.3.2 Ontology Creation and Data Annotation......Page 110
  5.3.3 Feature Extraction......Page 112
  5.3.4 Attribute Detection......Page 113
  5.3.5 Attribute Fusion with Low-Level Features......Page 114
  5.3.6 Attribute Selection and Weighting......Page 115
  5.4.1 Datasets......Page 117
  5.4.2 Attribute Analysis......Page 118
  5.4.3 Attribute Detection......Page 119
  5.4.4 Using Attributes to Re-identify......Page 120
  5.4.5 Re-identification with Optimised Attributes......Page 121
  5.4.6 Zero-shot Identification......Page 124
 5.6 Further Reading......Page 126
 References......Page 127
 6.1 Introduction......Page 130
 6.2 Related Work......Page 132
 6.3 Person Re-identification with by Latent SVM......Page 133
  6.3.1 Body Part Detection and Feature Representation......Page 134
  6.3.2 Definition and Estimation......Page 135
  6.3.3 Person Re-identification with by Latent SVM......Page 137
  6.4.1 The NUS-Canteen Database......Page 141
  6.4.2 Evaluation......Page 143
  6.5.1 Holistic Versus Part-Based Feature Representation......Page 144
  6.5.2 Prediction......Page 146
  6.5.3 SVM Versus Latent SVM......Page 147
 References......Page 148
 7.1 Introduction......Page 150
 7.2 State of the Art......Page 152
 7.3 Overview of our Approach......Page 153
 7.4 Details of our Approach......Page 155
  7.4.1 Part Detection......Page 156
  7.4.2 Pose Estimation......Page 160
  7.4.4 Feature Extraction......Page 161
  7.4.6 Multi-Shot Iteration......Page 163
 7.5 Training......Page 164
 7.6 Experiments......Page 165
 7.7 Conclusions......Page 169
 References......Page 170
8 One-Shot Person Re-identification  with a Consumer Depth Camera......Page 172
 8.1 Introduction......Page 173
 8.2 State of the Art......Page 174
 8.3 Datasets......Page 175
  8.4.1 Feature-Based Re-identification......Page 176
  8.4.2 Point Cloud Matching......Page 180
  8.5.1 Tests on the BIWI RGBD-ID Dataset......Page 184
  8.5.3 Multiframe Results......Page 188
  8.5.4 Runtime Performance......Page 189
 References......Page 190
9 Group Association: Assisting Re-identification by Visual Context......Page 193
 9.1 Introduction......Page 194
 9.2 Related Work......Page 196
  9.3.1 From Pixel to Local Region-Based Feature Representation......Page 197
  9.3.2 Center Rectangular Ring Ratio-Occurrence (CRRRO)......Page 198
  9.3.3 Block-Based Ratio-Occurrence (BRO)......Page 199
 9.4 Group Image Matching......Page 201
  9.5.1 Re-identification by Ranking......Page 202
  9.5.2 Re-identification with Group Context......Page 203
  9.6.2 Evaluation of Group Association......Page 205
  9.6.4 Improving Person Re-identification by Group Context......Page 208
 9.7 Conclusions......Page 209
 References......Page 210
10 Evaluating Feature Importance  for Re-identification......Page 212
 10.1 Introduction......Page 213
 10.2 Recent Advances......Page 214
 10.3 Feature Representation......Page 215
  10.4.1 Random Forests......Page 218
  10.4.2 Prototype Discovery......Page 221
  10.4.3 Prototype-Sensitive Feature Importance......Page 222
  10.4.5 Fusion of Different Feature Importance Strategies......Page 223
  10.5.1 Settings......Page 224
  10.5.2 Comparing Feature Effectiveness......Page 227
  10.5.3 Discovered Prototypes......Page 229
  10.5.4 Prototype-Sensitive Versus Global Feature Importance......Page 230
 10.6 Findings and Analysis......Page 234
 References......Page 235
Part IIMatching and Distance Metric......Page 238
 11.1 Introduction......Page 239
 11.2 BTF (Brightness Transfer Function)  Based Methods......Page 241
 11.3 Unsupervised Methods  for Collecting Training Data......Page 243
 11.4 Implicit Camera Transfer......Page 244
 11.5 An Explicit Camera Transfer Algorithm (ECT)......Page 246
  11.6.1 Explicit Versus Implicit Transfer Modeling......Page 247
  11.6.2 Camera-Dependent  Transfer-Based Versus Camera-Invariant Similarity-Based Methods......Page 248
 References......Page 251
12 Mahalanobis Distance Learning for Person Re-identification......Page 255
 12.1 Introduction......Page 256
  12.2.1 Mahalanobis Metric......Page 258
  12.2.2 Linear Discriminant Analysis......Page 259
  12.2.4 Information Theoretic Metric Learning......Page 260
  12.2.5 Large Margin Nearest Neighbor......Page 261
  12.2.6 Efficient Impostor-Based Metric Learning......Page 262
 12.3 Person Re-identification System......Page 263
  12.3.1 Representation......Page 264
 12.4 Re-identification Datasets......Page 265
  12.4.3 PRID 2011 Dataset......Page 266
  12.4.5 PRID 450S Dataset......Page 267
  12.5.1 Dataset Evaluations......Page 268
  12.5.2 Discussion......Page 272
 12.6 Conclusions......Page 273
 References......Page 274
13 Dictionary-Based Domain Adaptation Methods for the Re-identification of Faces......Page 276
 13.1 Introduction......Page 277
  13.1.1 Sparse Representation......Page 279
 13.3 Domain Adaptive Dictionary Learning......Page 280
 13.4 Unsupervised Domain Adaptive Dictionary Learning......Page 282
  13.4.1 Learning Intermediate Domain Dictionaries......Page 283
  13.4.2 Recognition Under Domain Shift......Page 284
  13.5.1 DADL for Pose Alignment......Page 285
  13.5.2 DADL for Face Re-identification......Page 287
  13.5.3 Unsupervised DADL for Face Re-identification......Page 288
 13.6 Conclusions......Page 290
 References......Page 291
 14.1 Introduction......Page 293
 14.2 Related Work......Page 295
 14.3 Identity Inference as Generalization of Re-identification......Page 296
  14.3.1 Re-identification Scenarios......Page 298
  14.3.2 Identity Inference......Page 299
 14.4 A CRF Model for Identity Inference......Page 300
 14.5 Experiments......Page 302
  14.5.1 Datasets and Feature Representation......Page 303
  14.5.2 Multishot Re-identification Results......Page 305
  14.5.3 Identity Inference Results......Page 308
 14.6 Conclusions......Page 311
 References......Page 312
 15.1 Introduction......Page 314
  15.2.1 Tracking without Using Appearance......Page 316
  15.2.2 Tracking with Sparse Appearance Cues......Page 319
  15.3.1 Detecting People Against a Static Background......Page 323
  15.3.2 Detecting People Against a Dynamic Background......Page 325
  15.3.3 Appearance-Free Experimental Results......Page 326
  15.4.1 Color Histograms......Page 328
  15.4.3 Face Recognition......Page 329
  15.4.4 Appearance-Based Experimental Results......Page 330
 15.5 Conclusions......Page 333
 References......Page 334
Part IIIEvaluation and Application......Page 336
 16.1 Introduction......Page 337
  16.2.1 VIPeR......Page 338
  16.2.2 i-LIDS......Page 339
  16.2.3 CAVIAR4REID......Page 340
  16.2.4 ETHZ......Page 341
  16.2.5 SARC3D......Page 342
  16.2.6 3DPeS......Page 343
  16.2.7 TRECVid 2008......Page 344
  16.2.10 QMUL Underground Re-identification (GRID) Dataset......Page 345
  16.2.12 RGBD-ID......Page 346
 16.3 Evaluation Metrics for Person Re-identification......Page 347
  16.3.1 Re-identification as Identification......Page 348
  16.3.2 Re-identification as Recognition......Page 349
  16.3.3 Re-identification in Forensics......Page 350
 References......Page 351
 17.1 Introduction......Page 354
  17.2.1 System Diagram......Page 355
  17.2.2 Low-Level Features......Page 357
  17.2.3 Semantic Features......Page 360
  17.2.4 Learning Feature Transforms Across Camera Views......Page 361
  17.2.5 Metric Learning and Feature Selection......Page 362
 17.3 Benchmark Datasets......Page 364
 17.4 Evaluation......Page 366
 17.5 Conclusions......Page 369
 References......Page 370
18 People Search with Textual Queries About Clothing Appearance Attributes......Page 374
 18.1 Introduction......Page 375
 18.2 Dissimilarity-Based Appearance Descriptors......Page 376
 18.3 A General Method for Implementing People Search  with Textual Queries......Page 379
  18.4.1 Implementation......Page 382
  18.4.2 Experimental Results......Page 385
 18.5 Conclusions......Page 390
 References......Page 391
 19.1 Introduction......Page 393
  19.1.1 Related Work......Page 394
 19.2 Detecting Camera Overlap......Page 396
  19.2.1 Mutual Information......Page 397
  19.2.3 Conditional Entropy......Page 398
  19.3.1 Calculating Cell Occupancy Probability......Page 399
  19.3.2 Camera Synchronisation......Page 400
  19.4.1 Ground Truth Comparison......Page 401
  19.4.2 Application to Re-identification......Page 405
  19.4.3 Scalability to Large Networks......Page 407
 19.5 Conclusions......Page 411
 References......Page 412
20 Scalable Multi-camera Tracking in a Metropolis......Page 414
 20.1 Introduction......Page 415
  20.2.1 Relative Feature Ranking......Page 417
  20.2.2 Matching by Tracklets......Page 418
  20.2.3 Global Space--Time Profiling......Page 419
  20.2.4 `Man-in-the-Loop\' Machine-Guided Data Mining......Page 420
  20.2.5 Attribute-Based Re-ranking......Page 422
 20.3 Implementation Considerations......Page 424
 20.4 MCT Trial Dataset......Page 427
  20.5.1 Associativity......Page 429
  20.5.2 Capacity......Page 433
  20.5.3 Accessibility......Page 434
 20.6 Findings and Analysis......Page 436
 References......Page 438
Index......Page 440