A Data-centric Cooperative Sensing Scheme in Crowdsourcing Systems

In a densely deployed crowdsourcing system, data observed at neighboring participants often exhibit strong spatial correlation. Exploiting this property, one may put a portion of participants into low power sleep mode without compromising the quality of sensing or the connectivity of the network. In this work, two fundamental scheduling questions are considered: (a) how to select a maximum number of participants to be put into sleep mode so that the overall sensing data integrity is maintained above a given threshold, and (b) how to divide the participants into two "shifts" such that participants at different shifts will perform sensing alternatively while sensing data integrity is being maximized. For question (a), we propose a novel data centric approach to explicitly exploit data correlation among participants. We formulate this subset selection problem as a constrained optimization problem and propose an efficient polynomial time algorithm. For question (b), we formulate this set partitioning problem as a constrained mini-max optimization problem. We validate these algorithms using the New Library of Wuhan University data set and observe very satisfactory results.