Adaptive crowd sensing with privacy-preserving WiFi fingerprinting

Abstract

This paper presents ongoing work in the context of a recently funded research project, MoniCrowd, aimed at advancing crowd monitoring in dynamic urban settings, particularly during temporary public events. The proposed system adopts an adaptive architecture based on passive Wi-Fi probe request detection, comprising portable sensors with multi-radio access connectivity and a rule-based fingerprinting method for anonymous and reliable device counting. Sensors autonomously select the best uplink from available connectivity options, enabling operation in suboptimal locations. A novel tool, the Information Elements Automatic Analyser (IEAA), enhances fingerprint robustness through fine-grained feature selection. A new dataset, collected under controlled Faraday cage conditions, supports this development. Preliminary field deployment results show strong correlation with manual counts, validating the approach under real-world conditions. The proposed fingerprinting method also achieved top accuracy in the international CONFRONT challenge. To optimise sensor performance in diverse environments, a UAV-assisted calibration tool is under development; its design and preliminary sensitivity mapping results are presented. Altogether, this work lays the foundation for scalable, autonomous crowd sensing solutions that can be rapidly deployed in dynamic urban contexts without requiring specialised local expertise.