Insights in the criticality of bicycle-car and bicycle-truck turning interactions at an urban intersection

Kurzfassung

Bicyclists represented 14.6% of the road fatalities in Germany in 2019. These accidents hap-pen mostly at urban intersections where bicyclists and motorized road users (MRU) encoun-ter each other. From all the types of MRU that bicycles interact with, trucks represented only 2% of the accidents with bicyclists between 2000 and 2014 in Norway, but up to 20% of road fatalities. Real bicycle-MRU interactions in the context of an urban intersection are studied here to get insights in the behavior of the actors and to identify the sources of risk for bicyclists. For this purpose, new traffic metrics to model criticality (Surrogate Measures of Safety or “SMoSs”) and evasive maneuvers in bicycle-MRU interactions are developed and validated. Scenarios including all four arms of the intersection, different combinations of lanes and also different road user types are considered. In order to gain an understanding of their key characteristics, the bicycle-truck interactions are compared to their bicycle-car equivalents. A recently developed collaborative scenario mining platform was used to automatically identify real traffic scenarios of bicycle-MRU interactions from trajectory data of the AIM Research Intersection in Braunschweig, Germany. This is a large-scale research facility, which records trajectory data with 20 fps with several stereo-cameras at a traffic signal-controlled crossing with bicycle paths. The trajectory data contains information about GNSS-based timestamp, location (UTM), velocity, acceleration, road user type (e.g., pedestrian, bicycle, car) and dimensions of the road users. A total of 196 hours of traffic were analyzed, with interactions occurring most frequently between 6 a.m. and 6 p.m. No interactions were analyzed after 8 p.m. and up to 6 a.m. Three main results are expected from this study of bicycle-MRU interactions. First, novel SMoS that are capable to detect critical and atypical situations as well as evasive maneuvers. Second, observation of patterns in the behavior and origins of conflicts. Third, the main dif-ferences between the bicycle-car and bicycle-truck interactions based on selected parame-ters. A part of the mined scenarios was labelled by human observers to annotate the criticali-ty, atypicality and observed patterns by reviewing the recorded camera data. The post-encroachment time (PET) is a popular SMoS to estimate criticality, but fails in case of interactions where the second road user arriving at the conflict point reacted by braking or swerving (false negative) or if it accelerated in a controlled situation (false positive) to close the gap. On the other hand, the new SMoSs do not incur in these errors because they consider several seconds of the maneuver before the crossing time. One of them was capable to find evasive maneuvers by comparing the projected PET (assuming constant velocities of both actors until the crossing point) during the maneuver with the final PET of the interaction. Bicycle-truck interactions were exceptionally rare in comparison to bicycle-car interactions. The main differences between both types of interactions were examined based on selected dynamic parameters as well as SMoS and the lateral deviation from the driving lane. It ap-peared that trucks deviated more from the driving lane than the cars, but their speeds and PET values were similar. Also, trucks had lower accelerations and decelerations. In conclusion, a better understanding of bicycle-MRU interactions at an urban intersection was obtained, behavioral patterns were detected and models of atypicality and criticality were implemented and validated. This is helpful, for example, for the scenario-based testing and development of automatic driving functions, achieving better traffic simulations and design of infrastructure.