Validating traffic models using large-scale Automatic Number Plate Recognition (ANPR) data

INTRoduCTIoN Automatic Number Plate Recognition (ANPR) entails the automated recording of the number plate, date/time and location of each vehicle that passes a roadside camera, using vehicle number plate recognition software. Records of individual vehicles that pass multiple cameras can be matched to determine the path of the vehicle and calculate travel times between the survey locations. If cameras are in a closed cordon, the origin and destination of external trips passing through the cordon can be determined. A series of ANPR cameras along a route, or at strategic locations throughout a network, would not observe every vehicle upon entry and exit to the network, and constitutes an open format number plate survey. Both closed and open format ANPR data have the potential to provide information that can be useful during the development of strategic traffic models, in ways that are not possible with other sources of traffic data. Comprehensive traffic observations from loop detectors, like ANPR, provide link speed and volume information which is useful during the calibration and validation of traffic models. But the additional ability of ANPR to track individual vehicles from point to point also provides potentially useful data on the distribution of trips through the network. While this constitutes partial rather than comprehensive origin-destination (OD) data, it may still serve as an additional independent data set against which model outputs can be validated. ANPR data has rarely been used in this way. The objective of this paper is to examine the use of ANPR data for traffic model validation in terms of its comprehensiveness and accuracy. ANPR data is provided by the South African National Roads Agency SOC Ltd (SANRAL) from the Open Road Tolling (ORT) system deployed on the Gauteng Freeway Improvement Project (GFIP). Selected link volumes and journey times are, for demonstration purposes, compared with the GFIP traffic model’s 2015 forecasts. In addition, the trip distribution characteristics of the ANPR data are exploited by extracting partial OD and trip length distribution metrics for comparison with modelled quantities. This required the development of a new methodology to process traffic model outputs such that they are directly comparable to ANPR-derived partial OD data. This is a feature of model validation that has not been found in previous studies. Validating traffic models using large-scale Automatic Number Plate Recognition (ANPR) data