Individual Tracking Using Multiple-Camera Under Non-Overlapping Environments

• Main Authors: Huang, Ding-Jie, 黃鼎捷
• Other Authors: Wang, Shuenn-Jyi
• Format: Others
• Language: zh-TW
• Published: 2011
• Online Access: http://ndltd.ncl.edu.tw/handle/77140656413716412298

碩士 === 國防大學理工學院 === 資訊科學碩士班 === 99 === With the prevalence of surveillance cameras, various kinds of intelligent video surveillance systems are developed. Signal-camera systems can be used to analyze simple human activities only in a small area. On the contrary, multi-camera systems can monitor widespread areas of activities such as tracking the route of an individual in a building, or mining the shopping behavior of a customer in a hypermarket. Multi-camera systems can be categorized to overlapping and non-overlapping environments. The advantages of non-overlapping environments are lower deploying cost, wider monitoring areas, and more flexible applications than those of overlapping environments. To track and record individual activities accurately, it is necessary to extract appropriate features from individuals. In this thesis, a visual tracking method using multiple cameras under non-overlapping environments is proposed to record and track individual activities in a building or hypermarket. First, a background subtraction scheme is used to detect moving individuals, and each individual is segmented to a trunk and a lower limb. Then, we extract color features and texture features from the trunk and the lower limb using the mean-shift algorithm and the adaptive LBP (ALBP) method, respectively. Finally, a feature-table and a camera-table are constructed to record each individual features and the detecting information of cameras, respectively, and we can trace each individual by analyzing these tables. The experimental results demonstrate that the proposed method can recognize individuals well, and the moving paths of each individual can be traced efficiently.