Detection and Classification of Bicyclist Group Behavior for Automated Vehicle Applications

Bicyclists are among the most vulnerable road users. Considering their unpredictable behavior in urban traffic scenarios, ensuring their safety is a complicated task. Automated vehicles are expected to interact and cooperate with vulnerable road users in the process of resolving complex traffic situations by maximizing traffic safety and traffic efficiency. In this context, correctly identifying and classifying individual bicyclists in group formations can provide an automated vehicle with an additional layer of information for the behavior of nearby bicyclists and enable safe interaction and communication strategies among automated vehicles and bicyclists in complex traffic scenarios. This paper presents a methodology for detecting and classifying bicyclists in groups or as a single bicyclist. The model is developed using trajectory data gathered at an unsignalized intersection in the city center of Munich, Germany. First, bicyclist trajectories are clustered using DBSCAN [1]. Then the trajectory similarity is evaluated using Discrete Fréchet [2]. Finally, the trajectories are classified in bicyclist groups using DBSCAN [1] according to their spatial and simultaneous similarity. The proposed approach shows relatively good results in identifying and classifying the bicyclist trajectories, while in post-analysis, correlations between group size and uniform bicyclist group behavior are identified.