Empirical study on pedestrian rotation mechanisms through bottlenecks.

We empirically investigated how pedestrians rotate through bottlenecks to avoid collisions. Shoulder data was found to be more reliable and accurate for measuring rotation compared to head trajectories. An angle exceeding 30^{∘} is used to identify the rotation state, with a false identification rate below 2.5%. Two types of rotation are observed: type I, where pedestrians actively rotate, gradually shifting their orientations away from the desired direction to adapt to confined space, and type II, where pedestrians rotate back. Statistical evidence indicates that the difference in blocking by opposite pedestrians and obstacles between the two sides of a square region in front of the pedestrian, is the potential mechanism triggering rotation behaviors, with a critical value of 20%. As blocking in that region and angular velocity increase, the rotation axis moves closer the pedestrian body center. The spatial distribution of rotation axes can be explained by the maximization of both short-term and long-term rotational yields. Additionally, in confined spaces, pedestrians need two or more step durations to complete the rotation, resulting in a longer rotation time. This paper enhances the understanding of the mechanisms behind human rotation through bottlenecks and provides empirical support for pedestrian rotation modeling.