Re-Identification of Trucks Based on Axle Spacing Measurements to Facilitate Analysis of WIM Accuracy

摘要

This paper examines weigh-in-motion (WIM) data from two different states to evaluate the performance of an improved re-identification methodology that has previously been used to match vehicles between WIM stations. The improvement allows the re-identification model to be trained without the use of ground truth data (i.e., true vehicle matches). The training data set is instead developed following a three step manual investigation of the characteristics of assumed vehicle matches between two WIM stations. The trained re-identification methodology was validated with data from Oregon, where the model was able to match identical vehicles with 70- 90% accuracy. The re-identification was then applied to data from two WIM stations in West Virginia where the downstream station had 6,178 vehicles with 10,427 possible matches at the upstream station. At a high confidence threshold of 0.98 (out of 1.0), the algorithm identified 526 likely matches. Additionally, this paper examines the differential calibration of the weight sensors at each WIM location by comparing the axle weight measurements between the matched vehicles. The data from the two WIM stations in Oregon illustrated good correlation in weight measurements. However, the two WIM stations in West Virginia did not illustrate consistent relationships across all axle measurements.