Your Search Path Tells Others Where to Park: Towards Fine-Grained Parking Availability Crowdsourcing Using Parking Decision Models

A main challenge faced by the state-of-the-art parking sensing systems is to infer the state of the spots not covered by participants’ parking/unparking events (called background availability) when the system penetration rate is limited. In this paper, we tackle this problem by exploring an empirical phenomenon that ignoring a spot along a driver’s parking search trajectory is likely due to the unavailability. However, complications caused by drivers’ preferences, e.g. ignoring the spots too far from the driver’s destination, have to be addressed based on human parking decisions. We build a model based on a dataset of more than 55,000 real parking decisions to predict the probability that a driver would take a spot, assuming the spot is available. Then, we present a crowdsourcing system, called ParkScan, which leverages the learned parking decision model in collaboration with the hidden Markov model to estimate background parking spot availability. We evaluated ParkScan with real-world data from both off-street scenarios (i.e., two public parking lots) and an on-street parking scenario (i.e., 35 urban blocks in Seattle). Both of the experiments showed that with a 5% penetration rate, ParkScan reduces over 12.9% of availability estimation errors for all the spots during parking peak hours, compared to the baseline using only the historical data. Also, even with a single participant driver, ParkScan cuts off at least 15% of the estimation errors for the spots along the driver’s parking search trajectory.