Design of an Adaptive Bayesian System for Sensor Data Fusion

Many artificial intelligent systems exploit a wide set of sensor devices to monitor the environment. When the sensors employed are low-cost, off-the-shelf devices, such as Wireless Sensor Networks (WSN), the data gathered through the sensory infrastructure may be affected by noise, and thus only partially correlated to the phenomenon of interest. One way of overcoming these limitations might be to adopt a high-level method to perform multi-sensor data fusion. Bayesian Networks (BNs) represent a suitable tool for performing refined artificial reasoning on heterogeneous sensory data, and for dealing with the intrinsic uncertainty of such data. However, the configuration of the sensory infrastructure can significantly affect the performance of the whole system, both in terms of the uncertainty of the inferred knowledge and in term of the hardware performance of the sensory infrastructure itself. This chapter proposes an adaptive Bayesian System whose goal is to infer an environment feature, such as activities performed by the user, by exploiting a wide set of sensory devices characterized by limited energy resources. The system proposed here is able to adaptively configure the sensory infrastructure so as to simultaneously maximize the inference accuracy and the network lifetime by means of a multi-objective optimization.