Real-time multi-sensor multi-source network data fusion using dynamic traffic assignment models

Abstract

This paper presents a model-based data fusion framework that allows systematic fusing of multi-sensor multi-source traffic network data at real-time. Using simulation-based Dynamic Traffic Assignment (DTA) models, the framework seeks to minimize the inconsistencies between observed network data and the model estimates using a variant of the Hooke-Jeeves Pattern Search. An empirical validation is provided on the Brisa A5 Inter-City Motorway in the West coast of Portugal. The real-time network data provided by loop detectors, video cameras and toll counters is collected and fused within DynaMIT, a state-of-the-art DTA system. State estimation is first performed, yielding consistent approximation of the network condition. This is then followed by network state forecast, showing significantly improved Normalized Root Mean Square Error (RMSN) over alternative predictive systems that do not use real-time information to correct themselves.