Bayesian Models for Reidentification of Trucks Over Long Distances on the Basis of Axle Measurement Data

Vehicle reidentification methods can be used to anonymously match vehicles crossing 2 different locations on the basis of vehicle attribute data. In this article, reidentification methods are developed to match commercial vehicles that cross 2 weigh-in-motion sites in Oregon that are separated by 145 miles. Using vehicle length and axle data as attributes to characterize vehicles, a Bayesian model is developed that uses probability density functions obtained by fitting Gaussian mixture models to a sample data set of matched vehicles. The reidentification model when applied to a test data set (where each downstream vehicle also crosses the upstream site) matches vehicles with an accuracy of 91% when both axle weight and axle spacings data are used. To account for the fact that not all vehicles in the downstream also cross the upstream site, an additional new step is developed to screen mismatched vehicles produced by the algorithm. For this step, several screening methods are developed that allow the user to trade off the total number of matched vehicles and error rate. For evaluating the effectiveness of the screening methods, 2 scenarios are considered. In the first scenario, only common vehicles that cross both the upstream and downstream sites are considered, whereas in the second scenario all downstream vehicles are considered. It is shown that the mismatch error can be reduced to as low as 1% and 5% at the expense of not matching about 25% of the common vehicles (crossing both sites) for the first and second scenarios, respectively.