Discrete choice models of pedestrian behavior

We propose a discrete choice framework for pedestrian dynamics, modeling short term behavior of individuals as a response to the presence of other pedestrians. We use a dynamic and individual-based spatial discretization, representing the physical space. We develop a model predicting where the next step of a walking pedestrian will be, at a given point in time. The use of the discrete choice framework is justified by its flexibility, the capacity to deal with individuals and the compatibility with agent-based simulation. The model is calibrated using data from actual pedestrian movements, manually taken from video sequences. We present two different formulations: a cross nested logit and a mixed nested logit. In order to verify the quality of the calibrated model, we have designed and developed a pedestrians simulator.