The simulation tools of people's displacements become more and more popular for applications emerging in the field of mobility planning, traffic management, impact assessment for city design and infrastructure modifications. Moreover, there is a lack of computational tools for the microscopic simulation of urban interactions between drivers and pedestrians. Feeling that road crossing is currently the main problem with pedestrian behavioural models, we conducted a laboratory experiment in order to understand to what extend the pedestrian's visual environment contribute to the crossing decision in order to improve a computational street crossing model. In the experiment, 36 12-second-video clips were presented to 32 participants, in conditions close to the crossing situation (scale 1, 160° of angle displayed on 6 large screens). The subjects were asked if they would have cross the street at the end of each clip. Two hypotheses were under investigation. The first one focuses on the objective description of the road crossing environment in terms of visual cues relevant for the crossing decision (traffic light, approaching vehicles, other pedestrians, etc.). The subject's answers were compared to the coding of the visual environment. The second hypothesis focuses on the subject's own explanations, about their motivations for crossing / not crossing. In both cases, the statistical analysis (logistic regressions) suggests that the crossing decision does not use the same visual cues depending of the presence/absence of traffic lights. The main result of this study is that the relevant visual cues are not the same at the signalized and at the unsignalized crossing, which leads to build separate quantitative models.
