Analyzing video sequences of multiple humans : tracking, posture estimation, and behavior recognition / Jun Ohya, Akira Utsumi, Junji Yamato.

Format:

Book

Author/Creator:

Ohya, Jun.

Contributor:

Utsumi, Akira.

Yamato, Junji

Rosengarten Family Fund.

Series:

Kluwer international series in video computing

Language:

English

Subjects (All):

Computer vision.

Physical Description:

xxii, 138 pages : illustrations ; 24 cm.

Place of Publication:

Boston : Kluwer Academic Publishers, [2002]

Contents:

2 Tracking multiple persons from multiple camera images / Akira Utsumi 7

2.2 Preparation 9

2.2.1 Multiple Observations With Multiple Cameras (Observation Redundancy) 9

2.2.2 Kalman Filtering 12

2.3 Features of Multiple Camera Based Tracking System 15

2.4 Algorithm for Multiple-Camera Human Tracking System 17

2.4.1 Motion Tracking Of Multiple Targets 17

2.4.2 Finding New Targets 21

2.5 Implementation 22

2.5.1 System Overview 22

2.5.2 Feature Extraction 23

2.5.3 Feature Matching at an Observation Node 26

2.6 Experiments 26

Appendix Image Segmentation using Sequential-image-based Adaptation 35

3 Posture estimation / Jun Ohya 43

3.2 A Heuristic Method for Estimating Postures in 2D 46

3.2.2 Locating significant points of the human body 46

3.2.2.1 Center of Gravity of the Human Body 46

3.2.2.2 Orientation of the Upper Half of the Human Body 48

3.2.2.3 Locating Significant Points 49

3.2.3 Estimating Major Joint Positions 52

3.2.3.1 A GA Based Estimation Algorithm 52

3.2.3.2 Elbow Joint Position 53

3.2.3.3 Knee Joint Position 53

3.2.4 Experimental Results and Discussions 54

3.2.4.1 Experimental System 54

3.2.4.2 Significant Point Location Results 54

3.2.4.3 Joint Position Estimation 54

3.2.4.4 Real-time Demonstration 57

3.3 A Heuristic Method for Estimating Postures in 3D 60

3.3.2 Image Processing for Top Camera 61

3.3.2.1 Rotation Angle of the Body 61

3.3.2.2 Significant Points 63

3.3.3 Estimating Major Joint Positions 65

3.3.4 3D Reconstruction of the Significant Points 66

3.3.5 Experimental Results and Discussions 67

3.3.5.1 Experimental System 67

3.3.5.2 Significant Point Detection Results 67

3.4 A Non-heuristic Method for Estimating Postures in 3D 70

3.4.2 Locating Significant Points for Each Image 70

3.4.2.1 Contour analysis 71

3.4.2.2 The tracking process using Kalman filter and subtraction image processing 73

3.4.3 3D Reconstruction of the Significant Points 77

3.4.3.1 Front image 77

3.4.3.2 Side image 77

3.4.3.3 Top image 78

3.4.3.4 Estimating 3D coordinates 79

3.4.4 Experimental Results 79

3.4.4.1 Experimental System 79

3.4.4.2 Experimental Results 80

3.5 Applications to Virtual Environments 86

3.5.1 Virtual Metamorphosis 86

3.5.2 Virtual Kabuki System 87

3.5.3 The "Shall We Dance?" system 92

4 Recognizing human behavior using Hidden Markov Models / Junji Yamato 99

4.2 Hidden Markov Models 102

4.2.2 Recognition 104

4.2.3 Learning 104

4.3 Applying HMM to time-sequential images 105

4.4 Experiments 108

4.4.1 Experimental conditions and pre-processes 108

4.4.2 Experiment 1 111

4.4.2.1 Experimental conditions 111

4.4.2.2 Results 111

4.4.3 Experiment 2 111

4.4.3.1 Experimental conditions 111

4.4.3.2 Results 112

4.5 Category-separated vector quantization 114

4.5.1 Problem in VQ 114

4.5.2 Category-separated VQ 114

4.5.3 Experiment 114

4.6 Applying Image Database Search 121

4.6.1 Process overview 121

4.6.2 Experiment 1: Evaluation of DCT 121

4.6.3 Experiment 2: Evaluation of precision-recall 124

4.6.4 Extracting a moving area using an MC vector 126

5 Conclusion and Future Work / Jun Ohya 133.

Notes:

Includes bibliographical references and index.

Local Notes:

Acquired for the Penn Libraries with assistance from the Rosengarten Family Fund.

ISBN:

1402070217

OCLC:

49284113