Research & Implementation Of Multi-sensor Information Fusion Platform In Smart Space

Rapid development of computer networks and multimedia technology has pushed the development of numerous E-Learning applications. As an important supporting technology, multisensor information fusion, which involves buiding intelligent space and designing fusion algorithm, is gaining increasing attention of researchers.At present, research related to buiding intelligent space is still in a talor-according-to-demand state. Research in modelization is in its infancy and no consented method is designed to build intelligent space with good extensibility rapidly. What is more, information hiding depends on complex cryptological algorithms, which are not flexible enough to use in real world.In the aspect of information fusion algorithm, most fusion algorithm cannot handle situations when there are large amount of sensor data. Performance drops greatly when sensor number increases.To address above issues, this paper researched on related theories and implemented a demo system, InfoIntegrator, on the basis of theoretical research.Work in this paper includes:1) Differences between different modeling methods of intelligent space are compared. Object-oriented intelligent space modeling method was proposed. This method can not only reuse ready system to the full extent, but also can meet the need to hide information on different secrecy level using encapsulation. Inherit mechanism is adopted to implement extensibility of intelligent spaces and space lists are used to connect spaces together. Using object oriented method, uniformed interface and outside behavior will be constructed rapidly on ready systems, greatly improving efficiency on building intelligent spaces.2) Multisensor information fusion algorithms were researched and the fusion algorithm based on rough set and neural networks was proposed. This algorithm used rough set theory to reduct input information of sensors and input reducted information to neural networks for further process. Performances are improved by reducing the amount of input data to neural networks. This algorithm not only takes advantage of the practicability of neural networks,but also avoids the headache of dimension disaster by reducing data dimensions using rough set reduction. 3) A prototype information fusion platform InfoIntegrator based on object orientation was designed. According to the modeling method proposed in this paper, focusing on the application of Shanghai Jiao Tong University's standard natural classroom, a platform was implemented to fuse multisensory information, enabling transparent communication among different sensor systems.There are two pending patents related to above work and several papers are accepted by key journals in China.