Liu, Jiachao (Carnegie Mellon University), Guarda, Pablo (Fujitsu Research of America), Niinuma, Koichiro (Fujitsu Research of America), Qian, Sean (Carnegie Mellon University)

Enhancing Multi-Class Mesoscopic Network Modeling with High-Resolution Satellite Imagery

Scheduled for presentation during the Poster Session "Detection, estimatation and prediction for intelligent transportation systems" (WeAT17), Wednesday, September 25, 2024, 10:30−12:30, Foyer

2024 IEEE 27th International Conference on Intelligent Transportation Systems (ITSC), September 24- 27, 2024, Edmonton, Canada

This information is tentative and subject to change. Compiled on December 26, 2024

Abstract

This study develops a novel framework to calibrate multi-class mesoscopic network models utilizing traffic density measurements derived from high-resolution satellite imagery. Satellite imagery provides city-wide traffic information that can compensate for the sparsity of the traditional data sources used to measure traffic conditions. To leverage this data, a computer vision pipeline is implemented to detect vehicles and match them to each link of a transportation network. Then, a dynamic origin-destination demand estimation (DODE) model is calibrated to reproduce the traffic density estimates derived from the computer vision pipeline. The loss function of the DODE model also includes components to jointly reproduce measurements of traffic counts and travel times and which may be available in multiple periods of the day. To assess the correctness and scalability of our framework, we conduct experiments on synthetic and real-world data. The results of out-of-sample tests in synthetic data demonstrate that incorporating density estimates with other traffic data can improve calibration accuracy, particularly for links lacking traffic sensor observations of traffic flow or travel time. The results with real-world data show that our method is scalable to large transportation networks.