A Generic Context Information System for Intelligent Vision Applications

The future intelligent vision is expected to be highly context-aware such that it can perceive and be aware of user's situation and react accordingly. In this paper, we propose a context representation mechanism and build a high-performance, extensible, distributed context information system based on it, in order to facilitate context-awareness development and information sharing. It pays attention to representing and organizing contexual information in an effective way and does not force any certain type of context reasoning algorithm. It can provide information-related services for distributed intelligent vision applications, mainly including representation, storing and retrieval, forming a whole pipeline of real-time semantic metadata generating and management. Besides user context, which is used to support runtime context communication between application components, our system also contains contextual descriptions about running environment and system configuration, making applications based on it can move to another environment or configuration seamlessly. Moreover, context representation in our system has a well-designed plugin-based architecture, helping users add their own context types without any modification of the original system. We introduce a context-aware meeting application based on our system, which employs Dynamic Bayesian Network as context reasoning algorithm. Experiment results show our context information system has excellent configurability, extensibility and performance.