Research on Road Traffic State Reconstruction Using License Plate Recognition Data

Abstract

Path flow estimation provides a detailed characterization of travel demand distribution in road networks and serves as a fundamental basis for transportation planning and management. Existing studies mainly rely on mobile sensor data, which are often sparse and limited by modeling assumptions, resulting in low estimation accuracy. To address this issue, this study proposes a path flow estimation method using license plate recognition (LPR) data as the sole input. The method is developed within a generalized least-squares framework by minimizing the weighted errors of path flows and turning flows. LPR data from 22 intersections in Kunming (248,798 records) during peak periods are used for validation. First, missing vehicle trajectories are reconstructed using particle filtering. Then, turning flows and their priors are estimated via tensor decomposition. A path–turning flow association matrix is constructed based on network topology to derive path flow priors, and an improved least-squares path flow estimation (ILS-PFE) model is solved using the conjugate gradient method. Results show that estimation errors remain stable when the sampling rate exceeds approximately 60%, but increase rapidly below this level. Thus, 60% is identified as an empirical reference threshold for LPR deployment. The proposed method demonstrates strong effectiveness and supports traffic state assessment, origin–destination (OD) inference, and signal control optimization.