Opportunistic Multiparty Calibration for Robust Participatory Sensing

Abstract

While bringing massive-scale sensing at low cost, mobile participatory sensing is challenged by the low accuracy of the sensors embedded in and/or connected to the smartphones. The mobile measurements that are collected need to be corrected so as to accurately match the phenomena being observed. This paper addresses this challenge by introducing a multi-hop, multiparty calibration method that operates in the background in an automated way. Using our method, sensors that are within a relevant sensing (and communication) range coordinate so that the observations of the participating (previously) calibrated sensors serve calibrating the other participants. As a result, our method is particularly well suited for participatory sensing within crowd meetings, as as for instance within public spaces. Our solution leverages multivariate linear regression, together with robust regression so as to discard the measurements that are of too low quality for being meaningful. To the best of our knowledge, we are the first to introduce a multiparty calibration algorithm, while previous work in the area focused on pairwise calibration. The paper further introduces a supporting prototype implemented over Android, and related experiment in the context of noise sensing. We show that the proposed multiparty calibration system enhances the accuracy of the mobile noise sensing application.

When the Power of the Crowd Meets the Intelligence of the Middleware

“…The middleware must as far as possible enhance locally the quality of the observations, from calibration to contextualization. While calibration may be achieved through regression analysis [19], contextualization requires prediction. The intelligent mid-dleware must implement soft/virtual sensors (as opposed to hardware/physical sensors) that run on the user device to further analyze and mine the data provided by the ever growing set of embedded cheap sensors.…”

Multi-Sensor Calibration Planning in IoT-Enabled Smart Spaces

“…Unfortunately, the proposed calibration protocol places a high demand on the end-users who must actively participate in the calibration. We subsequently investigated a distributed protocol for the opportunistic multi-party calibration of devices located in the same sensing and communication range [12].…”

“…In the case of mobile sensors, multi-hop calibration [22], [23], [24] allows mobile sensors to get calibrated "indirectly" using other mobile sensors that have been calibrated by the reference sensor. Multi-party calibration [12] derives regression models to solve the calibration problem when there are multiple (more than two) participants. Recent work also investigates planning problems, e.g., [24] explores the in-situ placement of reference sensors in the field; [7] proposes a TSP based path planning algorithm for a mobile calibrator.…”

“…The unnumbered constraints above are related to the definition of sensor selection and TTNC in Section III-A. Constraint (10) says the iterations need to cover the entire time span of T ;(11) and(12) make sure all the spots with selected sensors in Γ are visited in {W};(13) says no calibrator should work for longer thanĉ in any iteration. Additional constraints apply to ensure {W} are valid path(s).…”

Matching Technological & Societal Innovations: The Social Design of a Mobile Collaborative App for Urban Noise Monitoring

“…It follows that the collection of noise measurements through MPS delivers useful data mostly when used in a proactive data gathering mode (i.e., the Journey mode of the Ambiciti app). This further suggests to promote such use toward relevant stakeholders, as well as to investigate automated approaches to the correction of observation errors (e.g., see [22]). …”