Modelling driver behaviour in longitudinal vehicle-pedestrian scenarios - Analysis of driver comfort zone boundaries from naturalistic driving data and field tests

Abstract

In 2016, pedestrians accounted for the largest share within fatalities of vulnerable road users in traffic inside the European Union with a 21% fatality rate among all road users [1]. Active safety systems aim at avoiding crashes with pedestrians, however, there is a need to develop systems which are accepted by drivers, i.e. systems which are able to avoid collisions while keeping false alarm rate as low as possible. There has been a substantial amount of previous research with focus on driver interaction with pedestrians in crossing situations, but longitudinal scenarios have not gained the same amount of attention in research yet. This thesis aims at exploring the interaction between drivers and pedestrians in longitudinal scenarios, i.e. when drivers overtake or pass a pedestrian. To allowan analysis of driver behaviour, two data sets of realworld overtaking and passing manoeuvres were used. The first data set was extracted from the latest European naturalistic driving study, UDRIVE, and contained 77 overtaking events which occurred on rural roads in France. The second data set was collected within this thesis on a rural road in Sweden with a custom-made data logger. This second data set included 630 overtaking events which were collected by a pedestrian wearing the data logger. During the field test, experimental factors such as presence of oncoming traffic, walking direction of the pedestrian and lateral position of the pedestrian were varied. Results from the UDRIVE data set indicate that, in presence of oncoming traffic, drivers start to steer away earlier from a collision path than when oncoming traffic is absent. From field test data, a significant difference in minimum clearance between driver and pedestrian is shown for the factors oncoming traffic and walking direction. The implication is that drivers are willing to give less space to pedestrians in presence of oncoming traffic as well as when pedestrians are walking towards them. Almost 50% of the drivers in the field tests conducted in Sweden and about 90% of the french drivers from the UDRIVE data set kept less than 1.5 m distance to the pedestrian, less than the minimum distance set by policies in other European countries [2]. Hence, infrastructure especially in rural roads should be designed to allow safe and comfortable collision avoidance among road-users, especially in combination with pedestrians. Results for distribution of time to collision when steering away go in accordance to the corresponding Euro NCAP scenario. Bayesian linear regression was applied to model the event describing metrics minimum clearance and time to collision at the moment of steering away. Results show that the models are able to predict posterior distributions of those metrics as well as estimates and uncertainty of changes under the influence of certain factor combinations.