Bus passenger counts using Wi-Fi signals: some cautionary findings

Abstract

The viability of bus ridership surveys based on the detection of Wi-Fi MAC addresses of portable devices is analyzed. Motivation for the study arises from the apparent contradiction between success cases reported in the literature and empirical findings from field experiments we have carried out. Requirements for proper passenger identification in transit systems are used as the basis for evaluating the capabilities of commonly available detection hardware and software. More specifically, elapsed time intervals between detections of the same device are taken as the requirement for determination of the state of the device and, hence, the identification of the holder as a passenger. For instance, when performing boarding and alighting surveys with detection equipment placed onboard, it is necessary that multiple detections take place from right after passenger boarding and before he/she gets off, thus enabling accurate estimation of the trip origin and destination. Experimental results in controlled and uncontrolled trials indicate that off-the-shelf components used with available open source software may not grant successful detection. For instance, we have found times of up to 40 s for the first detection of 86% of nearby devices and an average of 80 s for a second detection of devices in the controlled experiment. For the uncontrolled experiment of rides on buses, significant differences between manual counts and detected devices were found. As a result of these empirical observations, careful assessment of the existing detection schemes used in ridership surveys is recommended.