Interdependence and Information Loss in Multi-Sensor Systems

摘要

often, information cues obtained by the sensors of the multi-sensor system are not completely independent. This is true since sensors operate in a close vicinity and are subjected to the same disturbances in the working environment. In some instances the observations made by the different sensors of the system are somewhat redundant. For example, there is always a possibility that a certain degree of overlap among these observations exists, and, hence, the estimates may be based, at least in part, on the same data. This problem could become more pronounced in intelligent sensors, where sensors are expected to possess some overlapping inferring ability, which in many cases leads to the possibility that two sensors or more base their decisions on similar underlying assumptions, theories, or common methods of analysis. Dependence in this case is not informative and therefore should be properly modeled so that its effect can be eliminated. Ignoring or improperly modeling this dependence may result in less informative observations and therefore sensors will tend to overestimate the importance of information cues communicated to them by other sensors. Consequently, the sensors may be perceived as being more accurate than they actually are. This article presents an information theory model for capturing dependence in multisensory data. A data fusion algorithm which revolves around the proposed dependency model model for minimizing the impact of dependence among sensors is also presented.