Location Privacy-Preserving Data Recovery for Mobile Crowdsensing

Abstract

Data recovery techniques such as compressive sensing are commonly used in mobile crowdsensing (MCS) applications to infer the information of unsensed regions based on data from nearby participants. However, the participants' locations are exposed when they report geo-tagged data to an application server. While there are considerable location protection approaches for MCS, they fail to maintain the correlation of sensory data, leading to the existence of unrecoverable data. None of the previous approaches can achieve both data recovery and data privacy preservation. We propose a novel location privacy-preserving data recovery method in this paper. Based on our discovery that the adjacency relations of non-zero elements are key to the missing data recovery in a crowdsensing data matrix, we design a correlation-preserving location obfuscation scheme to hide the participants' locations under effective camouflage. We also design an encrypted data recovery scheme based on the homomorphic encryption in order to avoid location privacy leakage from sensory data. Location obfuscation and data encryption preserve the participants' privacy, while the correlation-preserving and homomorphic properties of our method ensure data recovery accuracy. Evaluations of real-world datasets show that our privacy-preserving method can effectively obfuscate locations (e.g., yielding an average location distortion of 1.7km in a 2.4km x 4km area for successful location hiding), and it can efficiently achieve similar data recovery accuracy to compressive sensing (which has no privacy protection).