Towards Adaptive Sensory Data Fusion for Detecting Highway Traffic Conditions in Real Time

Abstract

The key challenge of detecting highway traffic conditions is to achieve it in a fully-covered, high-accuracy, low-cost and real-time manner. We present an approach named Megrez on the basis of treating mobile phones and probe vehicles as roving sensors, loop detectors as static sensors. Megrez can admit one or multiple types of data, including signaling data in a mobile communication network, data from loop detectors, and GPS data from probe vehicles, to carry out the traffic estimation and monitoring. In order to accurately reconstruct traffic conditions with full road segment coverage, Megrez provides a practical way to overcome the sparsity and incoherence of sensory data and recover the missing data in light of recent progresses in compressive sensing. Moreover, Megrez incorporates the characteristics of traffic flows to rectify the estimates. Using large-scale real-world data as input, we conduct extensive experiments to evaluate Megrez. The experimental results show that, in contrast to three other fusion methods, the results from our approach have high precisions and recalls. In addition, Megrez keeps the errors of estimates low even when not all three types of data are available.