Abstract:

The Smart Camera project at Princeton University aims at developing a flexible video analysis system with real-time performance. The core of the project is to develop a ubiquitous sensor framework for video surveillance applications.

The design framework presented in this dissertation consists of features from several levels in the embedded system design process. Features at the specification level include requirements analysis and categorization. Service-oriented modeling and migration methodology belong to the architecture level. Video analysis algorithms, software implementation and control authentication protocols are conducted at the component level. System verification is performed at the testing level.

Applications for ubiquitous sensors can be modeled as the composition of a system finite-state automaton, functional services and the middleware. There are two categories of designing tasks for the ubiquitous sensor systems: application-independent calibration and application-specific processing. A service-oriented software architecture allows the services to be dynamically reconfigured in response to changes in the system state.

In the dissertation, we present a prototype distributed gesture recognition system with true in-network processing and control. The system is equipped with two smart cameras with overlapped fields of view. It is capable of detecting and tracking people, as well as recognizing their activities in an application environment.