Estimation of Penetration Rates of Floating Car Data at Signalized Intersections

Abstract Floating Car Data (FCD) are a largely used data source, particularly well established in delivering real time and average measurements of speed and travel time. They have the advantage of a spread coverage in space and time, being totally independent from any field device deployment. In the same time, they present the weakness of incomplete vehicle coverages, and more importantly unknown penetration rates. This weakness is the main limiting factor toward full scale uses of FCD such as determining traffic volumes. The direct way to obtain the penetration rate of a FCD dataset and capture its variations is to compare FCD counts with counts from a stationary device. The method has a spatial limitation since the penetration rate can be known only in particular locations where a continuous full-count detector is available. This paper suggests a novel methodology to estimate an average penetration rate at signalized intersections based on the comparison between probe-aggregated congestion density and an average congestion density supposed known. Trajectories are first aggregated on a signal cycle basis. Probe macroscopic conditions are then measured from the aggregated plot. A fundamental diagram model fitting is employed to empirically measure the congestion density. The feasibility of the method is tested on a commercial dataset of unknown penetration rate. The relevance and accuracy of the method are then examined under different conditions and scenarios with randomly extracted trajectories from a microscopic simulation. The results suggest that the penetration rate when higher than 5% can be estimated with an error below 10% for a sampling rate up to 15 seconds.