A collaborative framework for semi-automatic scenario-based mining of big road user data

Traffic research has benefited from a significant expansion in the amount of available data. Consequently, the need arises for an automatic and efficient method to extract and analyze relevant traffic situations instead of a more traditional and manual approach like manual video annotation. This paper presents a framework to create such a data pipeline. The user must define the target scenarios and the pipeline will abstract the available trajectory data into candidate scenes (groups of interacting trajectories) and select the matches for the target scenarios. These scenes will be mined and modelled automatically for new valuable information. Furthermore, Surrogate Measures of Safety (SMoS) are applied to identify the critical and atypical scenes of the target scenarios. A set of eight scenarios containing interactions between bicycles and MRUs (Motorized Road Users) at the AIM (Application Platform for Intelligent Mobility) Research Intersection in the city of Braunschweig, Germany, was mined by a team of three researchers using the presented framework to validate it with positive results.