A Markovian model of pedestrian behavior

In this paper a statistical model of pedestrian behavior is proposed. This model is intended to be an important part of a study on the feasibility of car-to-pedestrian accident prediction. The proposed approach is phenomenological as it is based on a four discrete states Markov chain. These four states: "standing still", "walking", "jogging", and "running" are related to the pedestrian pace. First, given the former pedestrian state the current one is calculated. Then, the pedestrian speed vector is split up into the norm and the angle. Information on the statistical distribution of these quantities is available. Their values follow from the present pedestrian discrete state. The proposed model has been compared with related work. It has been used to generate statistically significant pedestrian trajectories and to predict car-to-pedestrian impacts. Simulation results are given based on an evaluation database of car and pedestrian accidents.