Network-Wide Traffic Volume Estimation Using Joint Matrix Factorization with Traffic Flow Conservation Law

Yuma Taguchi, Yoshinao Ishii, Takeyuki Sasai, Shintaro Fukushima, Katsushi Sanda

Abstract: Understanding urban traffic conditions, particularly the traffic volume on each road, is crucial in several applications, such as traffic planning, management, and environmental assessment. Conventionally, traffic volumes could only be observed on roads with traffic sensors installed. However, with the recent increase in connected vehicles, new problems for traffic volume estimation have emerged. Thus, we propose a network-wide traffic volume estimation method using data from connected vehicles with the observed traffic volume data. This method employs joint matrix factorization, which facilitates the integrated analysis of diverse data sources. Furthermore, we incorporated traffic flow conservation law to ensure the conservation of inflow and outflow at each intersection as a regularization term. We verified the proposed method using actual urban traffic data and confirmed that the proposed method achieved a higher accuracy than those of related baseline methods. Furthermore, we demonstrated the citywide validity of our estimation results by ensuring compliance with flow conservation law.