Multisensor data fusion and belief functions for robust singularity detection in signals

This paper addresses the problem of robust detection of signal singularity in hostile environments using multisensor data fusion. Measurement uncertainty is usually treated in a probabilistic way, assuming lack of knowledge is totally due to random effects. However, this approach fails when other effects, such as sensor failure, are involved. In order to improve the robustness of singularity detection, an evidence theory based approach is proposed for both modeling (data alignment) and merging (data fusion) information coming from multiple redundant sensors. Whereas the fusion step is done classically, the proposed method for data alignment has been designed to improve singularity detection performances in multisensor cases. Several case studies have been designed to suit real life situations. Results provided by both probabilistic and evidential approaches are compared. Evidential methods show better behavior facing sensors dysfunction and the proposed method takes fully advantage of component redundancy.