Urban Crowd Estimation via WiFi Probe Analysis

Nowadays, cities are places where data streams are continuously and ubiquitously exchanged between smartphones, IoT devices, and other modern technologies. The constantly growing urban populations make it necessary to provide stakeholders like transportation managers and security agencies with accurate methods for crowd-monitoring. In this way, more efficient space construction processes can be designed, i.e., by optimizing resource allocation or enhancing security measures. Analyzing network traffic has proven to be an effective method to estimate people’s presence in various specific areas. This thesis aims to provide a complete understanding of WiFi-based crowd-monitoring techniques, with their main advantages and drawbacks. We conducted an extensive experimental study of WiFi 802.11 probe requests, searching for new valuable information to enhance these systems’ performance and overcome their main limitations. The first phase of the research interested the content of the frames, particularly the Information Elements (IEs), fields that specify the characteristics of the source device. The findings were applied to ARGO, a network-based counting system developed in response to the Trialsnet European project. The conducted studies resulted in a new crowd-counting algorithm. It leverages additional fields of probe requests with respect to its predecessor and uses the OPTICS clustering algorithm to categorize and group Probe Requests based on the presumed source. The modifications applied resulted in an accuracy for crowd counting ranging from 83% to 93%. The results demonstrate the effectiveness of our adjustments, leading to an accuracy of up to 10 percentage points higher than the original. Additionally, the cluster quality has improved, as evidenced by higher values for homogeneity and completeness. In the second part of our work, we shift our focus to the time behavior of probe requests. We have examined how their sending rate varies according to the state of the channel, ranging from a non-congested state to a high-congested one. In the latter case, we observed a significant difference in the rates with respect to the ones recorded in isolated conditions. This analysis has shown that approaches used until now are not suitable, as they do not consider the impact of the congestion of WiFi channels in overcrowded environments. These observations have led to a new proposal for adapting the performance in diverse, specific contexts. This can lead to a more flexible algorithm but is feasible only when ground truth data are available, as the rate is tuned by minimizing the error relative to the effective ground truth. This research has shown that WiFi-based monitoring techniques offer a valid solution for crowd-estimation problems. The implemented system can provide accurate results, aligned with ground truth data. Furthermore, the affordability of this framework and its low power consumption make this system suited for daily monitoring applications, enabling a more efficient resource allocation and preserving the privacy of people in the monitored areas.