AbstractsComputer Science

Recently, there has been a growing interest in using microsimulation models for the safety assessment of road facilities by analyzing vehicle trajectories and estimating conflict indicators. Using microsimulation in safety studies can have several advantages. However, concerns have been raised about the ability of these models to realistically represent unsafe vehicle interactions and near misses and the need for a rigorous model calibration. The main objective of this thesis is to investigate the relationship between field-measured traffic conflicts and simulated traffic conflicts at signalized intersections. Automated video-based computer vision techniques were used to extract vehicle trajectories and identify field-measured rear-end conflicts. Conflict measures (e.g. time-to-collision (TTC)) and locations were determined and compared with simulated conflicts from the Surrogate Safety Assessment Model (SSAM) by analyzing the vehicles trajectories extracted from two microsimulation models: VISSIM and PARAMICS. To increase the correlation between simulated and field-measured conflicts, a two-step calibration procedure of the simulation models was proposed and validated. In the first calibration step, the simulation model was calibrated to ensure that the simulation gives reasonable results of average delay times. Then, in the second calibration step, a Genetic Algorithm procedure was used to calibrate the safety-related parameters in the simulation model. The correlation between simulated and field-measured conflicts was investigated at different thresholds of TTC. The results obtained from VISSIM and PARAMICS were compared. Furthermore, the transferability of the calibrated simulation models for safety analysis between different sites was investigated. As well, the spatial distributions of the field-measured and the simulated conflicts were compared through conflict heat maps. Overall, good correlation between field-measured and simulated conflicts was obtained after calibration for both models especially at higher TTC values. Also, the results showed that the simulation model parameters are generally transferable between different locations as the transferred parameters provided better correlation between simulated and field-measured conflicts than using the default parameters. The heat maps showed that there were major differences between field-measured and simulated conflicts spatial distribution for both simulation models. This indicates that despite the good correlation obtained, both PARAMICS and VISSIM do not capture the actual conflict occurrence mechanism.