Validating criticality for cyclist-vehicle interactions

Abstract

Ten interactions between cyclists and vehicles at a high-traffic intersection in Braunschweig, Germany were investigated. Trajectories, speed and distances in between road users were measured based on recordings from a high-end stereo-video camera system. As objective measures minimum time to collision (TTC_min), minimum time to zebra (TTZ_min), time integrated TTC (TIT), time exposed TTC (TET), minimum predictive encroachment time (PrET_min), pedestrian risk index (PRI) and an in-house metric, the Amberlight Risk Index (ARI) were calculated. As subjective measures the video recordings were rated for risk by three experts and 126 volunteers in an online study. A correlation analysis reveals that subjective ratings of volunteers correlate strongly with TTC_min and the maximum ARI and moderately with TET, TIT, TTZ_min and PRI. The expert ratings correlate moderately with PrET_min and weakly with TET, TIT and PRI. There was only a weak correlation between the ratings of the experts and volunteers. The TTC metric family (TTC_min, TTZ_min, TET and TIT) had strong correlations with each other. Based on subjective video ratings by humans as criterion, this work shows that the minimum of the popular TTC together with the ARI are well suited to predict criticality for cyclist-vehicle interactions at intersections as perceived by volunteers. However, the human ratings did not always agree. More research is needed to investigate why and which subjective risk rating is suited best for a given situation.