Improved Speed Estimates from Freeway Traffic Detectors

Abstract

This paper presents an analytical methodology to increase the accuracy of speed estimates from freeway traffic detectors by integrating information across lanes. Typically these estimates are quite noisy due to unobserved variables and measurement errors. Yet most traffic-monitoring applications employed by operating agencies only require a single measurement per direction at a detector station. Earlier efforts have focused on improving speed estimates of the individual lanes and then using a simple average to arrive at a single measure of speed at the detector station. But simply averaging the data across lanes is likely to preserve a measurement error in any one of the lanes. This paper improves speed estimates on a lane-by-lane basis using conventional aggregated flow and occupancy data obtained from single loop detectors. It then goes further by exploiting the information in the adjacent lanes to eliminate noise. In the course of analysis, data cleaning tools have been developed to identify and exclude malfunctioning detectors and transient errors, improving the validity of traffic data before estimating speed. These data cleaning tools consist of threshold value tests and basic traffic flow theory principles. To reduce the vulnerability to common transient measurement errors in individual lanes, the median of the improved estimates across lanes is taken in each direction. The methodology is then extended to clean noisy speeds measured from dual loop detectors and other sensors that mimic single loops.