Belief Revision in Multimodal Sensing and Detection Environments

In probabilistic data fusion systems, belief revision, i.e., the primary method for revising existing beliefs in response to new incoming evidence, is routinely implemented via the classic Bayes conditional. However, in complex multimodal data fusion environments, where data acquisition is often carried out via disparate sensors corresponding to multiple sensing modalities, implementation of traditional Bayes-like conditioning becomes a challenging task, primarily due to the complex nature of incoming evidence. In this article, a natural extension of Bayes-like conditioning applicable to belief revision in multimodal data fusion environments is explored. In particular, a detailed analysis of two functionally similar yet fundamentally different conditional approaches, namely, conditional update equation (CUE) and generalized conditional update (GCU), is presented. Furthermore, focusing on a particular set of multimodal belief revision problems (defined based on the relationship among sensing modalities), suitability of CUE and GCU for multimodal belief revision are discussed in light of two implementation strategies that are applicable to large-scale sensing and detection environments.