Top-view Trajectories: A Pedestrian Dataset of Vehicle-Crowd Interaction from Controlled Experiments and Crowded Campus

Predicting the collective motion of a group of pedestrians (a crowd) under\nthe vehicle influence is essential for the development of autonomous vehicles\nto deal with mixed urban scenarios where interpersonal interaction and\nvehicle-crowd interaction (VCI) are significant. This usually requires a model\nthat can describe individual pedestrian motion under the influence of nearby\npedestrians and the vehicle. This study proposed two pedestrian trajectory\ndatasets, CITR dataset and DUT dataset, so that the pedestrian motion models\ncan be further calibrated and verified, especially when vehicle influence on\npedestrians plays an important role. CITR dataset consists of experimentally\ndesigned fundamental VCI scenarios (front, back, and lateral VCIs) and provides\nunique ID for each pedestrian, which is suitable for exploring a specific\naspect of VCI. DUT dataset gives two ordinary and natural VCI scenarios in\ncrowded university campus, which can be used for more general purpose VCI\nexploration. The trajectories of pedestrians, as well as vehicles, were\nextracted by processing video frames that come from a down-facing camera\nmounted on a hovering drone as the recording equipment. The final trajectories\nof pedestrians and vehicles were refined by Kalman filters with linear\npoint-mass model and nonlinear bicycle model, respectively, in which\nxy-velocity of pedestrians and longitudinal speed and orientation of vehicles\nwere estimated. The statistics of the velocity magnitude distribution\ndemonstrated the validity of the proposed dataset. In total, there are\napproximate 340 pedestrian trajectories in CITR dataset and 1793 pedestrian\ntrajectories in DUT dataset. The dataset is available at GitHub.\n