About the Event
Air quality and personal pollutant exposure measurement are important for the health and productivity of individuals. Accurate measurement of personal exposure is challenging because of the spatially and temporally heterogeneous distribution of pollutant concentrations. We propose to use low-cost and miniature mobile sensor networks to provide real-time measurement of the environment directly surrounding the user. However, there are many challenges, including sensor drift, cross sensitivity, and outliers, to be addressed before mobile sensor network can be deployed in large scale and real-world applications.
My thesis aims to address those challenges by designing prototype sensor nodes of future generation mobile sensor networks, developing optimization techniques and systems, and evaluating the mobile sensor network in real-world deployments. My efforts can be generalized in four aspects: (1) we design the mobile sensor nodes and the mobile sensor network architecture that are capable of automatically collecting environment data and transferring them to the data base; (2) we model the sensor drift based on measurement and propose techniques such as collaborative calibration and optimal human mobility-aware sensor placement to minimize the drift error of individual sensors; (3) we model the pollutant concentration in indoor environment at the presence of inaccurate sensors. Based on the model, we develop hybrid sensor network synthesis technique to build accurate sensor networks under the constraint of budget; and (4) we propose a Bayesian network based sensor outlier detection and recovery system that can correct abnormal sensor readings, re-calibrate the sensor functions, and identify the gas composition is the environment simultaneously. All the techniques are evaluated and validated using the data collected from real-world deployment. Experiment and simulation results show that our collaborative calibration technique, hybrid sensor network synthesis technique, and outlier recovery technique can reduce the drift error significantly.