Analysis of partial registration plate data using a model with Poisson input and output

A popular method for obtaining an estimate of the amount of traffic making a particular movement is to conduct a survey in which the partial registration plates (typically the numerical part plus year letter) of vehicles passing a number of observation points are recorded. Although attempting to record only part of the plate significantly reduces the possibility of recording error, it creates a problem in the subsequent matching of data from different points, as vehicles are not uniquely identified by their partial registration number. There are a handful of statistical approaches which address this matching problem, but in practice it seems that ‘deterministic’ methods (such as those incorporated in most computer packages for analysing partial registration plate data) are often preferred. These make use of vehicle passage times in an attempt to exclude all ‘spurious’ matches. The aim of the work described in this paper is to develop a statistical framework for making use of passage times in a matching algorithm. A two observation point survey is considered, and a model proposed for vehicle arrivals in terms of time homogeneous Poisson processes. It is shown how, by exploiting well-known properties of such processes, both the journey time probability density function and the total number of true matches may be estimated. Practical aspects of the application of the algorithm are also discussed.