A bi-directional agent-based pedestrian microscopic model

This paper describes the development of a pedestrian microsimulation model that was developed based on the agent based modeling approach, which effectively accounts for the pedestrian intelligence and heterogeneity. The model focuses on producing accurate trajectories for pedestrian interactions. Behavior rules that control pedestrian interactions were extracted from a detailed pedestrian behavior study conducted in Vancouver, BC. The calibration of model parameters was performed using a Genetic algorithm, which aimed at minimizing the error between simulated trajectories and real trajectories obtained by means of computer vision. The validation of the results was conducted using two different data sets. The average errors between simulated and actual trajectories for the two data sets were 35 cm and 27 cm, respectively, while the average speed errors were 13.3% and 5.1%. Results also showed that the model was capable of predicting the correct collision avoidance strategy in 95% of the validation cases investigated.