Microscopic Estimation of Freeway Vehicle Positions Using Mobile Sensors

The introduction of mobile sensors, i.e. probe vehicles with GPS-enabled smart phones or connected vehicle technology, will potentially provide more comprehensive information on roadway conditions than conventional point detection alone. Several mobility applications have been proposed that utilize this new vehicle-specific data rather than aggregated speed, density, and flow. Because of bandwidth limitations of cellular and an expected slow deployment of connected vehicles, only a portion of vehicles on the roadway will be able to report their positions at any given time. This paper proposes a novel technique to analyze the behavior of freeway vehicles equipped with GPS receivers and accelerometers to estimate the quantity, locations, and speeds of those vehicles that do not have similar equipment. If an equipped vehicle deviates significantly from a carfollowing model’s expected behavior, the deviation is assumed to be the result of an interaction with an unequipped vehicle (i.e. an undetectable “ghost” vehicle). This unequipped vehicle is then inserted into a rolling estimation of individual vehicle movements. Because this technique is dependent on vehicles interacting during congestion, a second scenario uses an upstream detector to detect and insert unequipped vehicles at the point of detection, essentially “seeding” the network. An evaluation using the NGSIM US-101 dataset shows realistic vehicle density estimations during and immediately after congestion. Introducing an upstream detector to supply initial locations of unequipped vehicles improves accuracy in free flow conditions, thereby improving the root mean squared error of the number of vehicles within a 120-foot cell from 3.8 vehicles without a detector, to 2.4 vehicles with a detector, as compared to ground truth.