The inD Dataset: A Drone Dataset of Naturalistic Road User Trajectories at German Intersections

Automated vehicles rely heavily on data-driven methods, especially for\ncomplex urban environments. Large datasets of real world measurement data in\nthe form of road user trajectories are crucial for several tasks like road user\nprediction models or scenario-based safety validation. So far, though, this\ndemand is unmet as no public dataset of urban road user trajectories is\navailable in an appropriate size, quality and variety. By contrast, the highway\ndrone dataset (highD) has recently shown that drones are an efficient method\nfor acquiring naturalistic road user trajectories. Compared to driving studies\nor ground-level infrastructure sensors, one major advantage of using a drone is\nthe possibility to record naturalistic behavior, as road users do not notice\nmeasurements taking place. Due to the ideal viewing angle, an entire\nintersection scenario can be measured with significantly less occlusion than\nwith sensors at ground level. Both the class and the trajectory of each road\nuser can be extracted from the video recordings with high precision using\nstate-of-the-art deep neural networks. Therefore, we propose the creation of a\ncomprehensive, large-scale urban intersection dataset with naturalistic road\nuser behavior using camera-equipped drones as successor of the highD dataset.\nThe resulting dataset contains more than 11500 road users including vehicles,\nbicyclists and pedestrians at intersections in Germany and is called inD. The\ndataset consists of 10 hours of measurement data from four intersections and is\navailable online for non-commercial research at: http://www.inD-dataset.com\n