Data fusion for traffic flow estimation at intersections

Abstract

The efficient design and operation of intersections, particularly signalized intersections, depends on the availability of detailed traffic flow data. Information on the turning ratio and lane occupation is, however, commonly scarce or unreliable. Combining Floating Car Data (FCD) with other data sources offers the opportunity to fill this gap. As opposed to existing data fusion concepts, detector data is not directly used to deliver information on the traffic flow on a specific link, but it is evaluated in connection with the FCD to estimate the penetration rate of the traffic with probe vehicles. This information is combined with historic data and information contained in network connections. The data fusion itself is achieved by using a Kalman Filter (KF). By using FCD, the output of the data fusion process is up to date and can be used to take incidents and other fluctuations of traffic volumes into account. The elasticity of the filter to different input reliabilities and flow variations is shown. The concept works well and promises to be a useful tool for cities with sufficient FCD available. The performance depends in the first place on a good calibration, which can be achieved by extensive tests based on simulations and based on manual counts as part of an implementation.