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

Abstract

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.

Exploratory Methods for Truck Re-identification in a Statewide Network Based on Axle Weight and Axle Spacing Data to Enhance Freight Metrics

“…A recent paper summarizing some of the findings presented in this report recently was submitted for publication by the authors (Cetin et al 2010).…”

Reidentification of Trucks on Basis of Axle-Spacing Measurements to Facilitate Analysis of Weigh-in-Motion Accuracy

“…These methods rely on the variability of these attributes within the vehicle population and the ability to identify accurately the measurements that are generated by the same vehicle at both the upstream and the downstream stations. These measurements can be the derived physical attributes of the vehicle, such as length (4,5) and axle spacing (3,6,7), or some characteristics of the actual sensor waveform or inductive vehicle signature (8). Researchers have developed various methods, including lexicographic optimization (8) and Bayesian methods (3,7) to reidentify vehicles on the basis of these measurements.…”

“…These measurements can be the derived physical attributes of the vehicle, such as length (4,5) and axle spacing (3,6,7), or some characteristics of the actual sensor waveform or inductive vehicle signature (8). Researchers have developed various methods, including lexicographic optimization (8) and Bayesian methods (3,7) to reidentify vehicles on the basis of these measurements. In a typical implementation of these methods, a downstream vehicle is matched to the most similar upstream vehicle (or vice versa) on the basis of some defined metric (e.g., Euclidian distance).…”

“…This paper builds on its authors' previous work of using WIM data for reidentification by developing a systematic approach to creating a reliable data set for model training-calibration without the need for a ground truth data set. In previous studies, ground truth data (e.g., through unique transponder numbers or matching vehicle images from video) have been used to train a reidentification model (3,6,7). Such data may not be available or may be too costly to collect.…”

Software-Based Vehicle Reidentification with Existing Loop Infrastructure

“…The former include loops with high-speed scan detector cards, blade loops, or both (13)(14)(15)(16)(17)(18); electromagnetic sensor dots (19); and video. The latter include double loop (20)(21)(22)(23), single loop (24)(25)(26), or weigh-in-motion stations (27,28).…”