Abstract
Traffic monitoring devices count or classify vehicles or measure characteristics such as vehicle speed to portray traffic movement. The accuracy of these devices is often assessed by a comparison of aggregate or binned outputs with reference values. However, this process overlooks errors in the detection or measurement of individual vehicles. If vehicle-by-vehicle data are available, the accuracy analysis may be improved. Several methods are presented to assess traffic monitoring devices with vehicle-by-vehicle data. The approach taken is to use measures of agreement based on a frequency table, called a confusion matrix, which cross-classifies the data from traffic monitoring devices and a reference standard. In addition to the overall accuracy, several forms of the kappa statistic with and without weighting are presented. Prevalence is treated by normalization. Findings are illustrated by data from a study of automatic vehicle classifiers.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have