Adaptive turning flow estimation based on incomplete detector information for advanced traffic management

Modern traffic signal control systems call for reliable estimates of turning flows in real time to formulate effective control actions. This study proposes nonlinear least square (NLS) and extended Kalman filtering (EKF)algorithms to recursively estimate turning movement proportions in a network of intersections where only a partial set of detector counts are available. Using the large population approximation technique, a class of nonlinear, discrete-time Markovian traffic flow models are transformed into a linear state space model tractable for online applications. The quality of algorithms is evaluated with simulation data. As a comparison, the NLS algorithm shows less bias but with higher variance than the EKF algorithm.