A three-layer approach for protecting smart-citizens privacy in crowdsensing projects

A common approach in crowdsensing projects consists in using the sensing capabilities of mobile's user devices to collect some data, which is later aggregated and processed by a central collector, and then published in aggregated form. In return, individual users obtain useful information related to the provided data, making the exchange mutually beneficial. This use of crowds provides a framework to enable smart cities, as it channels a pervasive interaction between humans (smart-citizens) and multiple devices (sensors, terminals, and mobile phones, etc.) in order to cooperate and collaborate with their surrounding neighbors, thus building "smart" spaces where users can share data and enjoy services. In practice, things are never this simple. Smart-citizens may have reasons to question whether participating in these services is indeed convenient for them: in principle, their privacy can be easily vulnerated via mining the provided data. Moreover, recent revelations on massive surveillance by government agencies for security reasons only increase users' mistrust on smart-city services that need access to their sensitive data. In this article, we present a three-layer approach to secure the privacy of smart-citizens. Starting from the premise that raw data, if exists at all, should remain inside users' devices at all times and only aggregations must be collected by servers, we propose combining a first layer of homomorphic public-key encryption to secure locally stored data with a second layer that collect and compute a predefined set of aggregated statistics on user data in a secure environment using the homomorphic features of the encryption scheme. Finally, a third layer that adds differential privacy to control the dissemination of public information completes the approach. This way, citizens privacy is not compromised even in situations where naive release of the resulting aggregated data would have implicitly revealed information on a single user.