Vehicle path reconstruction using automatic vehicle identification data: An integrated particle filter and path flow estimator

Abstract

Automatic vehicle identification (AVI) can provide partial vehicle path data by matching the vehicle license plate on the detected links. However, the matched samples will rapidly degenerate with an increase in network size and a decrease in coverage rate and identification precision. In this paper, we propose an integrated macro–micro framework to reconstruct the complete vehicle path of realistic networks. The proposed framework integrates the individual path choice using particle filter (PF) at the microscopic level and the stochastic user equilibrium (SUE) principle with a path flow estimator (PFE) at the macroscopic level. The PF reconstructs the vehicle path by updating the state-space probability curve based on four observation models (i.e., path consistency model, AVI measurability criterion model, travel time consistency model and path attraction model) and incorporates a path flow constraint into the PFE model. The PFE minimizes the SUE objective while reproducing traffic counts on detected links and updates two of the four observation models (i.e., travel time consistency model and path attraction model) of the PF. The proposed method is tested on a realistic network for different AVI coverage rates ranged from 30% to 80%. The proposed method achieves approximately 55% improvement in link flow estimation and 67% improvement in path flow estimation compared with the original PFE without the microscopic level consideration. The accuracy of the vehicle path reconstruction exceeds 80% even when the AVI coverage is only 40% with an AVI detection error of 6%.