Two data sets related to pedestrian group identification work are shared.
- Synthetic data sets of pedestrian movement: The synthetic data sets simulate motion configurations of pedestrian movements, group formation and deformation. The data sets were recorded using a mouse tracking algorithm, where each continuous mouse movement mimics a pedestrian movement. Different crowd configurations are simulated with 2, 4, and 8 pedestrians in the video frame region. Motion configurations in the synthetic data sets are shown visually in the below figure. Each pedestrian is coded by an unique color for the trajectory. Synthetic data sets are created to check the clustering algorithm’s performance with different motion configurations occurring simultaneously and at different crowd densities. Combinations of these data sets, having 16, 32 pedestrians, were also used for performance evaluation. The synthetic data sets are available here .
- NUS Movie Event (NUSME) Data set: Most of the available data sets in the literature do not record the ground truth for group formation and movement. This data set records the pedestrians’ groups motion in the lobby area of a movie screening event at the National University of Singapore. The event attracted more than 200 visitors in 118 unique pedestrian groups, out of which 20 groups had their members occluded during some time of their existence. The region was video recorded from an aerial camera view point. The below figure shows the region with the location of the registration desk. The region has three points of entry / exit. In order to build ground truth without error of judgement, the visitors were requested to fill a form with the number of people in their group, time of visit and identification mark like shirt color.
A requisition form is shared for interested people to get access to the video data set and the ground truth of social groups in the pedestrian. The filled form should be sent to carun27(at)u(dot)nus(dot)edu for the data set access.