Models of pedestrian crossing behavior at signalized intersections

The traditional approach to reduce accidents at intersections has been to install traffic control devices. However, it has not been established that installation of such devices as signals, signs, or pavement markings, substantially improves pedestrian safety. On the contrary, empirical investigations have indicated that these devices tend to create a false sense of security. The major criteria used in installing these devices are vehicular volumes, pedestrian volumes, and engineering judgment. The compliance with these devices is, however, dependent on the pedestrian behavior. It is therefore necessary to study the crossing behavior, not only to realize the full benefits of signalization, but also to develop and evaluate new strategies to deal with the pedestrians. This report focuses on the crossing behavior of pedestrians at traffic signalised intersections. Since a non-compliant pedestrian attempting to cross on a "don't walk "phase looks for gaps in the traffic stream, the gap-acceptance theory is used to model the crossing maneuver. An inconsistent behavior model is assumed wherein the critical gap is treated as a random quantity varying both within and across individuals. Four possible crossing modes are identified, but only two could be studied because of sample size restrictions. Group interactions are incorporated in the models, as the behavior of individual pedestrians within the group may not be independent. Also, the pedestrians within the group may not be independent. Moreover, the pedestrian push-button choice behavior is integrated with the gap-acceptance models, as these two behaviors are correlated. Models are developed in the framework of random utility maximization theory using a multinomial probit approach. A data collection methodology was developed using a stratified sampling approach, with land use as the exogenous variable. The procedure was applied to selected intersections from the city of Austin, Texas. On-site surveys were conducted to obtain information on the behavior using a video recording technique. The calibration package used applies the Monte Carlo simulation technique to compute the choice probabilities, and obtains maximum likelihood estimates for the model parameters.