Evaluation of a signal state prediction algorithm for car to infrastructure applications

One of the most relevant pieces of information for cooperative driver assistance systems based on car to infrastructure communication is the upcoming signal state at signalized intersections. For example, functions like Green Light Optimized Speed Advisory need a short term forecast of the switching times. Unfortunately, in case of vehicle actuated control these switching times cannot be easily determined and need to be estimated using probability theory. In this paper different driver assistance functions and their respective requirements on the precision of the prediction are discussed. In order to assess the reliability of existing prediction methods, a well-known algorithm based on Markov chains is evaluated using a microscopic simulation study. The internal Markov states are represented by combinations of signal states and corresponding detector counts. In the model which is used for simulation two existing urban intersections are considered: Whereas one of the intersections is characterized by public transport priority, the signal times at the other intersection are adapted with respect to private cars only. Based on the results of the simulation study the performance and the deficiencies of the known Markov-chain based prediction method are analyzed in order to derive starting points for the development of an extended approach which takes additional constraints into consideration. Both, the known and the extended prediction method are then evaluated with respect to the specific requirements of the driver assistance functions. Based on the evaluation conclusions are drawn and recommendations for the application of the methods are given.