Abstract:
Vehicle classification is a vital measure used to ensure appropriate roadway design as it affects both capacity and pavement endurance. Given that, departments of transpo...Show MoreMetadata
Abstract:
Vehicle classification is a vital measure used to ensure appropriate roadway design as it affects both capacity and pavement endurance. Given that, departments of transportation across the US collect vehicle miles travelled (VMT) for their highways using automatic vehicle classifiers (AVC), and then use these figures for future highway design. Accuracy assessment of AVCs is thus important to ensure proper VMT reporting. Studying the accuracy of AVC devices is therefore essential. Previous studies employed either manual counting or a “play and pause” method of traffic video recording to verify the accuracy of AVC devices. This paper details a custom vision-aided software developed to aid in extracting accurate vehicle count and classification information used as ground truth data. Authors discuss the methodology used to study vehicle classification accuracy of AVC and weigh-in-motion sites tasked with vehicle classification. Several indicators introduced to investigate the accuracy of each site are highlighted. Results of a year-long 2013 study indicate a good performance of AVC devices and that the main source of error was the misclassification of class 2 and 3 vehicles as class 5.
Published in: 2014 IEEE Intelligent Vehicles Symposium Proceedings
Date of Conference: 08-11 June 2014
Date Added to IEEE Xplore: 17 July 2014
Electronic ISBN:978-1-4799-3638-0
Print ISSN: 1931-0587