Maximum likelihood estimation of ground truth for air quality monitoring using vehicular sensor networks

Abstract

Various works on vehicular sensor networks (VSNs) for air quality monitoring use solid-state gas sensors due to its low cost and compact form factor. However, solid-state gas sensors have poor selectivity and are sensitive to ambient temperature and relative humidity. In addition, the sensitivity and accuracy of solid-state gas sensors degrade over time due to aging effects. Frequent recalibration of these sensors are required to maintain the accuracy of their measurements. In large VSNs, it is impractical to manually calibrate each node. Thus, calibration must be performed automatically and in-field. Assuming that the gas concentration is homogenous within an area, co-located VSN nodes can either: (1) copy measurements from a highly accurate fixed station in their immediate vicinity, or, in the absence of a fixed station, (2) collaboratively estimate the ground truth. In this work, we use maximum likelihood estimation for determining the ground truth gas concentration in an area by fusing information from co-located sensors in a VSN. Through simulations, we show that the absolute errors of the proposed method has lower mean and standard deviation as compared with existing work.