Design and analysis of a real-time video human gesture recognition system
Wayne H. Wolf, Tiehan Lu · 2004
Video analysis systems are the systems developed to identify interesting objects, people, and recognize a certain set of activities. Such systems have been widely used in military, law enforcement, traffic management and other applications. The Smart Camera Project in Princeton University aims at developing a flexible video analysis system that can be used in wide range of video surveillance applications. The core of the Project is the development of a real-time video human gesture recognition system called SmartCamera. SmartCamera is able to detect and track interested objects. It can also detect people and recognize their activities in an application environment, such as in a room, plane, car, or security checkpoint. The results of SmartCamera analysis can also be used to control the operation of other devices in these environments. Designing a real-time video analysis system is a complex task, as many factors including processing speed, system cost, accuracy, and robustness, need to be carefully balanced. This dissertation addresses these factors at three levels, algorithm level, software level, and hardware level. Two new background elimination algorithms are proposed in the dissertation to enhance the performance of SmartCamera systems in changing background and varying lighting condition environment. The software implementation of the SmartCamera systems is then taken up, including the selection of platforms and the optimization process that leads to a 5-times speedup. New hardware architecture is then proposed that can provide more processing power than current platforms. The new architecture exhibits more than 10 times average speedup over a traditional architecture.