Large-scale congestion analysis using compressed measurements

Traffic congestion is a major concern in metropolitan areas and a quick congestion assessment of large-scale network is required for modern traffic management referred to as Intelligent Transport Systems (ITS). However, ITSs are facing the challenge of real-time storage, retrieval and processing of a vast amount of collected data over a large-scale network. Compressed sensing (CS) is an efficient technique of sampling high-dimensional data and is successfully used to reconstruct signals via a small set of non-adaptive, linear measurements. Whilst CS has mainly been applied for data reconstruction, we propose in this paper the use of CS for classification and estimation of some meaningful parameters in the traffic problem. To this end, we develop a novel sensing matrix for congestion analysis and show that for traffic data collected in Melbourne urban network, our CS provides an opportunity to extract the desired information from a small number of random projections. In addition, our method could improve the efficiency of database query and has the ability to deal with missing or defective data.