Abstract
This paper presents a design of an architecture that facilitates the work of a fusion engine. The logical combination or merging of input streams invoked by the fusion engine is based upon the definition of a set of patterns and its similarity with previously collected data from various modalities. Previous fusion engine designs had many weaknesses, among them their being specialized on a specific domain of application. The proposed architecture addresses such weakness and provides additional features, namely its ability to handle large number of modalities and due to its using knowledge base and standardization characteristics; it becomes suitable to various types of multimodal systems. The techniques used to achieve these features are discussed in this paper.