Model-Based Visual Tracking : The OpenTL Framework

Model-Based Visual Tracking : The OpenTL Framework

  • ただいまウェブストアではご注文を受け付けておりません。 ⇒古書を探す
  • 製本 Hardcover:ハードカバー版/ページ数 302 p.
  • 言語 ENG
  • 商品コード 9780470876138
  • DDC分類 006.37

Full Description


This book has two main goalsof this growing field, as well as to propose a corresponding software framework, the OpenTL library, developed by the author and his working group at TUM-Informatik. The main objective of this work is to show, how most real-world application scenarios can be naturally cast into a common description vocabulary, and therefore implemented and tested in a fully modular and scalable way, through the defnition of a layered, object-oriented software architecture.The resulting architecture covers in a seamless way all processing levels, from raw data acquisition up to model-based object detection and sequential localization, and defines, at the application level, what we call the tracking pipeline. Within this framework, extensive use of graphics hardware (GPU computing) as well as distributed processing, allows real-time performances for complex models and sensory systems.

Table of Contents

Preface                                            xi
1 Introduction 1 (11)
1.1 Overview of the Problem 2 (4)
1.1.1 Models 3 (2)
1.1.2 Visual Processing 5 (1)
1.1.3 Tracking 6 (1)
1.2 General Tracking System Prototype 6 (2)
1.3 The Tracking Pipeline 8 (4)
2 Model Representation 12 (43)
2.1 Camera Model 13 (13)
2.1.1 Internal Camera Model 13 (3)
2.1.2 Nonlinear Distortion 16 (1)
2.1.3 External Camera Parameters 17 (1)
2.1.4 Uncalibrated Models 18 (2)
2.1.5 Camera Calibration 20 (6)
2.2 Object Model 26 (13)
2.2.1 Shape Model and Pose Parameters 26 (8)
2.2.2 Appearance Model 34 (3)
2.2.3 Learning an Active Shape or 37 (2)
Appearance Model
2.3 Mapping Between Object and Sensor 39 (4)
Spaces
2.3.1 Forward Projection 40 (1)
2.3.2 Back-Projection 41 (2)
2.4 Object Dynamics 43 (12)
2.4.1 Brownian Motion 47 (2)
2.4.2 Constant Velocity 49 (1)
2.4.3 Oscillatory Model 49 (1)
2.4.4 State Updating Rules 50 (2)
2.4.5 Learning AR Models 52 (3)
3 The Visual Modality Abstraction 55 (23)
3.1 Preprocessing 55 (2)
3.2 Sampling and Updating Reference 57 (2)
Features
3.3 Model Matching with the Image Data 59 (11)
3.3.1 Pixel-Level Measurements 62 (2)
3.3.2 Feature-Level Measurements 64 (3)
3.3.3 Object-Level Measurements 67 (1)
3.3.4 Handling Mutual Occlusions 68 (2)
3.3.5 Multiresolution Processing for 70 (1)
Improving Robustness
3.4 Data Fusion Across Multiple 70 (8)
Modalities and Cameras
3.4.1 Multimodal Fusion 71 (1)
3.4.2 Multicamera Fusion 71 (1)
3.4.3 Static and Dynamic Measurement 72 (5)
Fusion
3.4.4 Building a Visual Processing Tree 77 (1)
4 Examples Of Visual Modalities 78 (84)
4.1 Color Statistics 79 (14)
4.1.1 Color Spaces 80 (5)
4.1.2 Representing Color Distributions 85 (4)
4.1.3 Model-Based Color Matching 89 (1)
4.1.4 Kernel-Based Segmentation and 90 (3)
Tracking
4.2 Background Subtraction 93 (3)
4.3 Blobs 96 (16)
4.3.1 Shape Descriptors 97 (7)
4.3.2 Blob Matching Using Variational 104(8)
Approaches
4.4 Model Contours 112(14)
4.4.1 Intensity Edges 114(5)
4.4.2 Contour Lines 119(3)
4.4.3 Local Color Statistics 122(4)
4.5 Keypoints 126(14)
4.5.1 Wide-Baseline Matching 128(1)
4.5.2 Harris Corners 129(4)
4.5.3 Scale-Invariant Keypoints 133(5)
4.5.4 Matching Strategies for Invariant 138(2)
Keypoints
4.6 Motion 140(7)
4.6.1 Motion History Images 140(2)
4.6.2 Optical Flow 142(5)
4.7 Templates 147(15)
4.7.1 Pose Estimation with AAM 151(7)
4.7.2 Pose Estimation with Mutual 158(4)
Information
5 Recursive State-Space Estimation 162(35)
5.1 Target-State Distribution 163(3)
5.2 MLE and MAP Estimation 166(6)
5.2.1 Least-Squares Estimation 167(1)
5.2.2 Robust Least-Squares Estimation 168(4)
5.3 Gaussian Filters 172(8)
5.3.1 Kalman and Information Filters 172(1)
5.3.2 Extended Kalman and Information 173(3)
Filters
5.3.3 Unscented Kalman and Information 176(4)
Filters
5.4 Monte Carlo Filters 180(12)
5.4.1 SIR Particle Filter 181(4)
5.4.2 Partitioned Sampling 185(2)
5.4.3 Annealed Particle Filter 187(2)
5.4.4 MCMC Particle Filter 189(3)
5.5 Grid Filters 192(5)
6 Examples Of Target Detectors 197(17)
6.1 Blob Clustering 198(4)
6.1.1 Localization with 199(3)
Three-Dimensional Triangulation
6.2 AdaBoost Classifiers 202(2)
6.2.1 AdaBoost Algorithm for Object 202(1)
Detection
6.2.2 Example: Face Detection 203(1)
6.3 Geometric Hashing 204(4)
6.4 Monte Carlo Sampling 208(3)
6.5 Invariant Keypoints 211(3)
7 Building Applications With Opentl 214(37)
7.1 Functional Architecture of OpenTL 214(2)
7.1.1 Multithreading Capabilities 216(1)
7.2 Building a Tutorial Application with 216(24)
OpenTL
7.2.1 Setting the Camera Input and 217(3)
Video Output
7.2.2 Pose Representation and Model 220(4)
Projection
7.2.3 Shape and Appearance Model 224(3)
7.2.4 Setting the Color-Based Likelihood 227(5)
7.2.5 Setting the Particle Filter and 232(3)
Tracking the Object
7.2.6 Tracking Multiple Targets 235(2)
7.2.7 Multimodal Measurement Fusion 237(3)
7.3 Other Application Examples 240(11)
APPENDIX A POSE ESTIMATION 251(14)
A.1 Point Correspondences 251(8)
A.1.1 Geometric Error 253(1)
A.1.2 Algebraic Error 253(1)
A.1.3 2D-2D and 3D-3D Transforms 254(2)
A.1.4 DLT Approach for 3D-2D Projections 256(3)
A.2 Line Correspondences 259(2)
A.2.1 2D-2D Line Correspondences 260(1)
A.3 Point and Line Correspondences 261(1)
A.4 Computation of the Projective DLT 262(3)
Matrices
APPENDIX B POSE REPRESENTATION 265(16)
B.1 Poses Without Rotation 265(3)
B.1.1 Pure Translation 266(1)
B.1.2 Translation and Uniform Scale 267(1)
B.1.3 Translation and Nonuniform Scale 267(1)
B.2 Parameterizing Rotations 268(4)
B.3 Poses with Rotation and Uniform Scale 272(3)
B.3.1 Similarity 272(1)
B.3.2 Rotation and Uniform Scale 273(1)
B.3.3 Euclidean (Rigid Body) Transform 274(1)
B.3.4 Pure Rotation 274(1)
B.4 Affinity 275(2)
B.5 Poses with Rotation and Nonuniform 277(1)
Scale
B.6 General Homography: The DLT Algorithm 278(3)
Nomenclature 281(4)
Bibliography 285(10)
Index 295