Improving the Accuracy of Vehicle Reidentification Algorithms by Solving the Assignment Problem

Vehicle attributes (e.g., length, sensor signature) collected at upstream and downstream points can be used to reidentify individual vehicles anonymously so that useful quantities such as travel times and origin–destination flows can be estimated. In typical reidentification algorithms, each downstream vehicle is matched to the most “similar” upstream vehicle on the basis of some defined metric. However, this process usually results in matching one upstream vehicle to more than one downstream vehicle, and some upstream vehicles are not assigned to any downstream vehicles. This paper presents a two-stage methodology to alleviate this problem, first by developing a Bayesian method for matching the most similar vehicles and then by defining and solving an assignment problem to ensure that each vehicle is matched only once. The results indicate that the proposed method, when applied to the sample field data collected by automatic vehicle classification and weigh-in-motion sensors, reduces the mismatch error by 15% to 60% and by an overall average of 42%. For the sample data, vehicles are matched with 99% accuracy after the methodology presented here is applied.