Research Team: Stephen Ritchie (lead), Kate Kyung Hyun, and Andre Tok
UC Campus(es): UC Irvine
Problem Statement: In response to the significant growth of freight movement in recent years, transportation agencies are increasingly aware of the need for detailed information on truck flows within highway network systems for traffic monitoring and operations, freight demand analysis and environmental impact studies. However, the limited availability of truck data sources complicates the capture of truck travel patterns which significantly vary by season, time of day and location.
Project Description: This study developed a methodology to accurately estimate network-wide truck flows by leveraging existing point detection infrastructure, namely inductive loop detectors. The tracking model identifies individual trucks at detector locations using advanced inductive signatures and matches vehicle pairs at detector locations, using an extended form of the Bayesian classification model to estimate matching and non-matching probabilities of the vehicle pairs. Several vehicle feature selection and weighting methods including Self Organizing Map and K-means clustering were applied to better identify individual vehicles from signature data. It was shown that the proposed extensive feature processing enhanced vehicle identification performance even among vehicle pools sharing similar physical configurations. The developed model was tested along an approximately 5.5-mile freeway segment on I-5 and CA-78 in San Diego, California where only 67 percent of the total trucks were observed at both up- and down-stream detector sites. Results showed balanced performances in exactness and completeness of matching with 91 percent of correct outcomes for multi-unit trucks.
Project Partner(s): California Air Resources Board, Caltrans