Real-Time Magnetic Length-Based Vehicle Classification: Case Study for Inductive Loops and Wireless Magnetometer Sensors in Oklahoma State

Abstract

Providing reliable, real-time vehicle volume and classification information is vital for 21st century intelligent transportation systems (ITS). Large vehicles have a major impact on traffic flow, as well as road maintenance. Information about passenger car versus truck volume is crucial for all transportation agencies. The work presented here provides a comprehensive method to develop length-based vehicle classification (LBVC) techniques that could be implemented using inductive loops (IDL) or magnetometer sensors (MAG). Distinctive LBVC schemes were developed to bin vehicles into groups based on structural similarity and statistical characteristics. Data collection, including vehicle magnetic length (VML) estimates using IDL and MAG, was performed at different sites located on Oklahoma highways and rural roadways to capture various traffic characteristics. Video images were utilized as ground-truth for accurate data labeling. Extensive data analyses, including machine learning methods and probabilistic modeling, were conducted to define decision boundaries for developed LBVC schemes. Three scenarios were developed for determining optimal thresholding methods: total classification accuracy maximization, per-group classification error minimization, and equal classification error optimization. Evaluation revealed consistent and accurate performance for all developed schemes. Classification accuracies of 97.70% and 99.00% were reported using MAG and IDL, respectively. The developed classification models are computationally efficient and can provide real-time LBVC data. The models are intended to supplement or replace axle-based data collection methods used throughout Oklahoma. The methodology developed in this work will also benefit other states and territories interested in developing LBVC schemes.