HUMAN APPEARANCE MODELING IN VISUAL SURVEILLANCE

We present an appearance model for establishing correspondence between tracks of people which may be taken at different places, at different times or across different cameras. Illumination insensitive color features, i.e., RGB rank feature and brightness-color feature are used. Path-length feature is added for structural information and invariance to motion and pose. The appearance model is constructed by kernel density estimation. Kullback-Leibler distance measures the similarity between the models. To further exploit the information in video sequence, key frame selection method and online hierarchical clustering algorithm are proposed to construct appearance model from video…

Contents

Introduction
2 Human Appearance Modeling from Snapshot
2.1 Overview
2.2 Review of existing human appearance models
2.3 Color path-length profile
2.3.1 Path-length
2.3.2 Color feature
2.3.2.1 RGB Color
2.3.2.2 Normalized RGB Color
2.3.2.3 RGB Rank
2.3.2.4 Brightness and Color Decomposition
2.4 Appearance Model
2.4.1 Appearance Model Using Kernel Density Estimation
2.4.2 Matching of Appearances
2.5 Evaluation
2.5.1 Bandwidth Selection
2.5.2 Experiment Setting
2.5.3 Study of Colo r- Path length Profile
2.5.4 Study of Scale and Sub sampling
2.5.5 Local Appearance Difference
2.6 Conclusion
3 Human Appearance Modeling of Video Sequence
3.1 Overview
3.2 Key Frame Selection
3.2.1 Algorithm
3.2.2 Experiment Results
3.3 Online Clustering
3.3.1 Algorithm
3.3.2 Experiment Results
3.4 Online Hierarchical Clustering
3.4.1 Algorithm
3.4.2 Experiment Results
3.5 Conclusion
4 Conclusion
Bibliography

Author: Yu, Yang

Source: University of Maryland

Download URL 2: Visit Now