Stochastic real-time urban traffic state estimation: searching for the most likely hypothesis with limited and heterogeneous sensor data

摘要

The aim of this thesis is to improve road traffic information for the purpose of traffic signal control. Loop detectors are the prevalent sensors for state-of-the-art adaptive traffic control systems. Loops provide accurate measurements, but have a limited vehicle detection area. Additional sensor installations increase the total detection area and can offer performance improvements for adaptive control. New sensor types become available from time to time and are often economically attractive candidates for additional installations. They provide measurements that are generally not compatible with existing loop detector measurements. The problems with adding new sensors are: (a) the new sensor's incompatibility is an integration challenge for existing systems and (b) the increased detection area is possibly still limited in total size and non-contiguous.The research challenge is to combine in real-time the data from loops and other sensors from a limited total detection area and estimate the traffic state in a useful form for adaptive traffic control. This thesis proposes an urban traffic state estimation method that performs data fusion of heterogeneous traffic measurements from a non-contiguous detection area that exploits vehicle interactions. The proposed method uses a Bayesian algorithm that searches for the most likely set of vehicle trajectories that justify the measurements. The most likely set of vehicle trajectories is found by combining (a) a reversible jump Markov chain Monte Carlo algorithm, (b) driver and vehicle behaviour modelling, and (c) probabilistic sensor modelling.The methodology is tested with different sensor types in simulated and field experiments and evaluated against ground truth trajectories. The result is a set of estimated vehicle trajectories updated in real-time that are measurement-consistent. A single sensor provides measurements of a specific type for a specific location on the road. In contrast, the estimated trajectories allow traffic engineers to extract most types of traffic information from anywhere on the road. This is achieved by programming virtual sensors that extract desired information from the estimated trajectories. The abstraction offered by the trajectories provides flexibility and a future-proof approach to new traffic sensing technologies.