Kalman Filter Applications for Traffic Management

Traffic congestion is a major problem in urban areas that has a significant adverse economic impact through deterioration of mobility, safety and air quality. As a result, the importance of better management of the road network to efficiently utilize existing capacity is increasing. To that end, many urban areas build and operate modern Traffic Management Centers (TMCs), which perform several functions, including collection and warehousing of real-time traffic data, as well as utilization of this data for various dynamic traffic control and route guidance applications. In order to be effective, these applications —which include Advanced Traveler Information Systems (ATIS) and Advanced Traffic Management Systems (ATMS)— require traffic models that provide, in real-time, estimation and prediction of traffic conditions. The complexity of transportation systems often dictates the use of detailed simulation-based Dynamic Traffic Assignment (DTA) models (Ben-Akiva et al., 1991, 2002, Mahmassani, 2001) for this purpose. Dynamic Traffic Assignment (DTA) systems support both planning and real–time applications. Planning applications may include the off–line evaluation of incident management strategies, the evaluation of alternative traffic signal and ramp meter operation strategies and the generation of evacuation and rescue plans for emergencies (e.g. natural disasters) that could affect the traffic network. Real–time applications make use of the traffic prediction capabilities of DTA systems and may include on–line evaluation of guidance and control strategies, real–time incident management and control, support of real–time emergency response efforts and optimization of the operation of TMCs through the provision of real–time predictions. Real–time DTA systems typically comprise two main functions: traffic state estimation, and traffic prediction (Ben-Akiva et al., 2002). An overview of the state-of-the-art Dynamic Traffic Assignment framework is shown in Fig. 1. DTA functionality is supported by two main modules: a demand simulator and a supply simulator. The demand simulator fuses surveillance information with historical information for the estimation and prediction of the evolving demand patterns. This is achieved through a combination of aggregate predictive