Using classification to improve wireless sensor network management with the continuous transferable belief model

摘要

We show how the performance of a sensor management algorithm can be improved by using the continuoustransferable belief model (cTBM). Bayesian approaches can have problems modelling uncertainty whereas thetransferable belief model (TBM) has been proven effective in dealing with this the TBM achieves this byassigning support to sets of events rather than just singleton events. The discrete nature of such set theoreticuncertain reasoning approaches (including Dempster-Shafer approaches) can have problems modelling continuoussignals such as the speed of a target; the cTBM has been developed to overcome such inherent problems. Existingwork at Cardiff University classifies targets by combining the cTBM and a particle filter; each particle is usedto construct a set of beliefs, which are then fused with the existing beliefs for classification-this is then usedto update the particle filter. Williams et al. provide a framework for managing sensor networks by balancingthe quality of information gained from a sensor network with the required communications cost when tracking atarget. Our proposed new system integrates the above approaches, and has similar basic communications costs to the latter. It is now not only able to track a target but also classify it the combination results in improvedperformance; this is shown in our simulation results from Monte Carlo trials with various scenarios.