Learning Multi-Sensor Confidence using Difference of Opinions

Multiple sensors are being employed in different environments for performing various observation tasks and detecting events of interest occurring in the environment. However, all the sensors deployed in the environment do not have the same confidence level due to their differences in capabilities and imprecision in sensing. The confidence in a sensor represents the level of accuracy that it provides in accomplishing a task, which can be computed by comparing the current observation of the sensor through tedious physical investigation. Confidence computed in this manner is static and does not evolve over time. Moreover, performing physical investigation for checking the accuracy of the sensor observation is not feasible in a running system due to the overhead in incurs. Nevertheless, it is essential to know how the sensors are performing in a real-time scenario. This paper addresses this issue and proposes a novel method to dynamically compute the confidence in sensors by learning the differences of their individual opinions with respect to the particular detection task. Experimental results show the suitability of using the dynamically computed confidence as an alternative to the accuracy measures of the sensors.