Highlights • We assess the smartphone probe accuracy for real-time traffic data. • We examine how accuracy is affected by road and traffic conditions. • A relationship is established between smartphone speed and GPS localization error. • The CEP range is only part of the uncertainty problem facing traffic estimates. • Smartphone probe estimates are useful for traffic management and safety analysis.

Abstract Before smartphone probes can be used to obtain instantaneous vehicle speeds and other dynamic characteristics, the accuracy of these estimates needs to be established under varying degrees of satellite signal interruption commonly found with varying road and traffic conditions. This paper presents the results of several vehicle tracking tests comparing smartphone speeds to benchmark values obtained for three types of routes. Benchmark values were obtained using a high frequency calibrated V-Box mounted on the test vehicle with four Android OS smartphone units. A relationship is established linking error in smartphone instantaneous speeds to the corresponding signal Circular Error Probable (CEP) range for different road and traffic conditions. This relationship is used to provide speed adjustment factors for the smartphone probe estimates subject to varying satellite signal strength. The CEP test is a reported GPS unit indicator of precision based on a known ground control benchmark. Smartphone speeds (adjusted and unadjusted) are compared to aggregate speed profiles from a stationary radar detector placed at a given location along SS106 in southern Italy. The smartphone devices were found to replicate closely the observed speed profiles obtained from the fixed detector station. Simple t-tests suggest that the means of the smartphone speeds for the unadjusted case differed significantly from the means obtained from the radar detector, when the smartphone estimates were adjusted for uncertainty (CEP range related), however, the difference in mean speeds between the smartphone probes and the radar detector profile was not found to be statistically significant.