The rounD Dataset: A Drone Dataset of Road User Trajectories at Roundabouts in Germany

Abstract

The development and validation of automated vehicles involves a large number of challenges to be overcome. Due to the high complexity, many classic approaches quickly reach their limits and data-driven methods become necessary. This creates an unavoidable need for trajectory datasets of road users in all relevant traffic scenarios. As these trajectories should include naturalistic and diverse behavior, they have to be recorded in public traffic. Roundabouts are particularly interesting because of the density of interaction between road users, which must be considered by an automated vehicle for behavior planning. We present a new dataset of road user trajectories at roundabouts in Germany. Using a camera-equipped drone, traffic at a total of three different roundabouts in Germany was recorded. The tracks consisting of positions, headings, speeds, accelerations and classes of objects were extracted from recorded videos using deep neural networks. The dataset contains a total of six hours of recordings with more than 13 746 road users including cars, vans, trucks, buses, pedestrians, bicycles and motorcycles. In order to make the dataset as accessible as possible for tasks like scenario classification, road user behavior prediction or driver modeling, we provide source code for parsing and visualizing the dataset as well as maps of the recording sites. For non-commercial public research, the dataset is available free of charge at https://www.round-dataset.com