Multi-sensor data fusion and reliable multi-channel computation: unifying concepts and techniques

Multiple sensors are used for much the same reasons as redundant computation channels; viz to tolerate certain types of error, incompleteness, inaccuracy, and/or tardiness in the data supplied by individual system elements such as sensors or processors. Data fusion in multi-sensor systems is investigated by numerous signal processing researchers. The problem of "voting" or, more generally, integrating results from multiple identical or diverse computations has been studied by computer scientists/engineers interested in fault tolerance and distributed computing issues. With very few exceptions, these groups publish their results in disjoint conferences and journals. Thus, there is little interaction and cross-fertilization between the two efforts. We review common aspects of the two problems and show how a data-driven methodology may offer potentials for tackling both problems within a unified framework.