Estimating Transit Ridership Using Wi-Fi Signals: An Enhanced Rule-Based Approach

This paper focused on the estimation of bus ridership using Wi-Fi probes (i.e., signals) emitted by smartphones that bus passengers carried. Smart stations -- which consist of a Raspberry-Pi computer, a Wi-Fi adapter, and a GPS add-on-- were programmed to sniff Wi-Fi signals and transmit signal data through a cloud service to the research computer. These smart stations were mounted onboard a network of transit buses that serve the City of Bozeman, Montana, and its surrounding areas. Two rule-based methods were developed to estimate the number of passengers onboard a bus at any given time. In the first, standard method, a signal was labeled as a passenger if it met arbitrary cutoff values from six criteria pertinent to speed (how fast the signal/device was traveling relative to the bus), duration being detected (as a proxy for how long the device remained in close proximity of the bus), and signal strength (which may correlate with the distance between the device and the bus). The second method employed a cost-function minimization via grid-search to tune the cutoff values involved in those subjective rules (e.g., a valid passenger signal should be close enough to a bus stop when it is first and last detected, but how close is close?). Results suggest a strong linear relationship between model-estimates and ground-truth passenger counts -- on average, the model estimates were able to capture 67% of the observed passenger counts. As Wi-Fi enabled personal devices continue to saturate the market, a Wi-Fi based counting tool as studied here can serve as an efficient way to monitor passenger flows of transportation systems.