A behavior-based methodology for evaluating pedestrian-vehicle interaction at crosswalks

This dissertation explores the interaction of pedestrians and drivers at unsignalized crosswalks in an event-based data collection and analysis approach. Through logistic regression techniques the microscopic data are used to derive predictive models for driver yielding and pedestrian crossing behavior. The analysis found that pedestrian and driver decision-making processes are sensitive to the dynamic profile of the vehicle, pedestrian characteristics and other concurrent events at the crosswalk. By relating the yield outcome to the dynamic state of the vehicle, a region of vehicle dynamics constraints was defined where virtually no yields are observed. Pedestrian assertiveness was found to be a key variable for promoting yielding behavior and increasing the likelihood of a pedestrian crossing. A contrast of the behavioral models for driver yielding and pedestrian crossing found a generally better model fit for the latter category. It is reasoned that the pedestrian decision is strongly influenced by the temporal duration to the point of conflict and the consequences of a poor decision. With a lack of enforcement, a driver is more easily swayed in the decision of whether or not to yield. The pedestrian crossing data are thus more consistent than the yielding data, resulting in models with greater statistical power. The evaluation of two pedestrian crossing treatments found that the treatments resulted in expected increases in the likelihood of drivers yielding, but also promoted more aggressive pedestrian behavior. For a pedestrian-actuated treatment, the effect on yielding was found to be greater following activation. The predictive models for two geometrically different mid-block crossings showed sufficient similarities, which suggests that the development of generic yielding and crossing models is feasible. The results of this research demonstrated the viability of the data collection approach and gave promise for expandability of the method to other applications. The research is meaningful in the context of modern microsimulation models, where the resulting models can be applied to describe the interaction of the driver and pedestrian modes.