Abstract
Traffic flow sensors are crucial for transportation management and planning, but their accuracy is often compromised by various malfunctions. This study develops a robust framework for detecting traffic sensor malfunctions by leveraging lane-to-lane traffic correlations, addressing limitations of previous methods that focused solely on individual sensor performance. We introduce the coefficient of variation (CV) as a measure of lane-to-lane correlation, calculated as the ratio of standard deviation to mean of traffic measurements (flow rate or density) across lanes. Using the NGSIM dataset with detailed vehicle trajectories, we establish a range of CV values for accurately functioning sensors through hypothesis testing, based on Edie’s generalized definition of traffic variables. We then apply this CV range to identify malfunctioning loop detectors in the California PeMS dataset on State Route 91. This is based on the assumption that lane-to-lane correlations should fall within a specific range for correlated general-purpose lanes. In this dataset, density is calculated by dividing occupancy by the length of a loop detector and adding the vehicle length. Our methodology enhances the accuracy of detecting traffic sensor malfunctions, contributing to improved transportation management and planning. Future work could explore applying this approach to different road types and traffic conditions, including lanes with special purposes.