摘要: |
Vehicle re-identification matches vehicles crossing two different locations. Building on a previous study, we investigate different methods for re-identification and explore factors that impact accuracy. Archived data from weigh-in-motion (WIM) stations in Oregon are used to develop, calibrate, and test vehicle re-identification algorithms. In addition to the Bayesian approach developed in the previous study, a neural network model is developed. The results show both methods to be effective while the Bayesian method results are more accurate. A comprehensive analysis employing the Bayesian algorithm matches vehicles that cross upstream and downstream pairs of WIM sites. Data from 14 different pairs of WIM sites are used to evaluate how factors such as distance, travel time variability, truck volumes, and sensors impact accuracy. The testing showed a large variation in accuracy, with sensor accuracy and volumes have the greatest impacts, while distance alone shows no significant impact. |