Identifying Pedestrian Travel Patterns using Wi-Fi sensing technology

Abstract

Advances in information technology have provided opportunities to better understand urban activities and human mobility patterns for policy-makers and urban planners, enabling real-time and large-scale application in high resolution. Wi-Fi traces are one of the sources for detecting Wi-Fi equipment as it sends out packets of so-called probe requests to search for nearby networks. The probe request includes the MAC address, allowing us to identify each device by the movement trajectory of its user. This Wi-Fi tracking has gained attention recently with the mass introduction of Wi-Fi devices such as smartphones. Many researchers have suggested various ways of counting the number of pedestrians and analyzing their trajectories using the Wi-Fi sensing. However, novel methods are needed for identifying pedestrians' travel patterns. We can separate them into two categories: just passing by, or staying for a while on a street. Not only are these two behaviors different, but their relationship with surrounding environments needs to be determined. In this paper, a prototype based on the Raspberry Pi is set up to gather the Wi-Fi probe request on the UNIST campus. Algorithms to filter and aggregate data collected by the active scanning devices and spatial visualizations of Wi-Fi usage on the campus are investigated. Several feature-values are employed for classifying the users’ behavior patterns: average, maximum, and minimum speed, variance of speed, and time stayed. Some measurements driven by algorithms developed for GPS tracking, i.e., stay point detection, are covered. Extracting the features from each segment, K-mean clustering method is used to classify the pedestrians' travel patterns. The results are verified with the manual labelling of the ground truth for a small number of devices. Rather than the count of pedestrians, we explore the significant potential of Wi-Fi probe request by identifying pedestrians’ travel patterns. These recognized patterns describe better the characteristics of the streets and spaces. For example, the number of people who stayed on the streets is more directly related to their vitality compared with the number of people passing a certain point we've been using so far. The efforts to promote these patterns for improving the street vitality can be addressed immediately as this Wi-Fi sensing is being generated in real-time based on flexible location and environment. Without additional infrastructure investments, this work can rapidly be extended to already existing public Wi-Fi networks. It is expected that the relationship with other environmental factors such as temperature and noise can be investigated when we concurrently obtain them from connected multiple sensors to the Wi-Fi scanner.