Federal Highway Administration Vehicle Classification from Video Data and a Disaggregation Model

Abstract

To assess highway system performance, states are required to submit compositional traffic count reports to the Federal Highway Administration (FHWA). A methodology was developed for obtaining compositional vehicle counts from low-cost portable video technology. Autoscope, an advanced video camera with machine vision processing, was used to capture, process, and store compositional counts based on vehicle length. A maximum of five length-based vehicle classes can be differentiated from video; therefore a methodology was developed that generates the FHWA 13-category vehicle classes. A disaggregation process, typically used in stochastic hydrology, is employed. Results from the disaggregation model are compared with actual FHWA classification data obtained from axle counters. It was observed that less than 1% of the vehicles were not included in the model output and 3.43% were misclassified. In addition, 95 and 99% confidence intervals calculated by the disaggregation model included 92.3 and 100% of the actual axle counter data.