Pedestrian Trajectory Data Receives International Attention
There are constantly new studies and reports coming out about autonomous and connected vehicles and how they interact with other vehicles, traffic lights, and even the roads they are driving on via underground cables, but it is equally important to understand how these vehicles interact with the pedestrians they come in contact with.
Sponsored by the U.S. Department of Transportation, Carnegie Mellon Mobility21 National University Transportation Center, a team of faculty and students led by Emeritus Professor, Umit Ozguner at Ohio State’s Center for Automotive Research have developed a new dataset that can help researchers better understand vehicle-pedestrian interaction in crowded areas. The dataset was created by collecting data from two scenarios. The first being an experimental setup where specific pedestrian-vehicle interaction patterns were created and a second where natural scenarios occurred on the crowded campus of Dalian University of Technology in Dalian, China. Data from both scenarios were captured by drones.
“Instead of analyzing the behavior of one or two pedestrians, our dataset considers the behavior of multiple pedestrians, i.e., crowd behavior,” said Dongfang Yang, a graduate student in the Department of Electrical and Computer Engineering. “Typical scenarios are shared spaces such as squares, train stations, shopping malls, university campuses, and old small streets, where the vehicle-pedestrian interaction is unavoidable. The new dataset can benefit the design of more realistic simulation models for vehicle-pedestrian interaction. It can also be used to improve the accuracy of pedestrian behavior prediction in autonomous driving.”
The dataset is publically accessible and has received the attention of several researchers, particularly from a European project, SocialCars. One researcher from the prooject, Fatema Johora, from Clausthal University of Technology in Germany has joined the team at Ohio State to work on a collaborative project about modeling multi-pedestrian behavior in mixed traffic scenarios. The new dataset will be utilized to validate the proposed methods.