Privacy-Preserving Crowd Estimation Using Multiple Wi-Fi Sensors

Wi-Fi-enabled devices frequently transmit probe request frames to locate nearby access points. These probe requests can be used to estimate the number of devices (and thus people) in an area and possibly their flow. This works irrespective of whether or not mobile devices randomize their MAC addresses, which not all do (sufficiently often). Covering larger areas with many sensors causes challenges because frames may be picked up by multiple sensors, resulting in double counting in such overlap areas. Duplicates could be removed by an aggregation server that collects reports from all sensors, but this would require sensors to reveal the (temporary) identifiers, i.e., MAC addresses, to the collection server. This paper presents a decentralized approach to removing duplicates using private set intersections and to aggregate crowd size reports from multiple sensors on a server in a privacy-preserving manner. We present our design and carry out a simulation-based evaluation comparing our crowd count estimates to ground truth.