Real-time queue estimator for signalised intersections - based on detector data

Kurzfassung

With the real-time queue estimator (EQ) it is possible to easily estimate a queue length at a traffic light (TL) in real time. The queue length could be estimated very precisely up to the most distant sensor with the aid of real-time information about TL sensors (e.g. induction loops, infrared detectors) and TL signal state. The queue length information can be used to calculate a leading time (green) required to reduce the queue length for e.g. approaching public transport or emergency vehicles. This in turn enables precise intervention in the TL control to give priority to vehicles while at the same time minimising the disruption this causes to the general traffic flow. Using the queue length for a GLOSA (Green Light Optimal Speed Advisory) service is another interesting possible area of application, as this additional information can be used to improve the GLOSA advice. The EQ can also be used to determine the number of vehicles in a road section. The EQ contains a simple traffic model based on a cellular automaton. In these automate, the movements of the vehicles (approaching the junction) are simulated. The vehicles are then either removed (if green light) or added to a virtual queue (if red light or the queue has not yet been removed). The concept of EQ is developed in python code and prototypically tested with Traffic Control Interface in conjunction with a SUMO-simulation. SUMO served as a source of input data for the detector data and signal state and offers the possibility of validation through the SUMO-GUI. It is possible to map all different infrastructure conditions, such as lane widening or lane addition or subtraction and the different availability as well as position of detectors and whether a traffic light status is available plus such as different traffic demands. False detections (no detection or double detection) of vehicles could also be caused in the SUMO-simulation and thus mechanisms could be implemented and tested to correct/intercept these. In a next step, this approach will also be tested at a real intersection with real input data.