Research Mentor(s): Neda Masoud, Assistant Professor
Research Mentor School/College/Department: Civil and Environmental Engineering, College of Engineering
Presentation Date: Thursday, April 22, 2021
Session: Session 4 (2pm-2:50pm)
Breakout Room: Room 20
Connected Vehicle (CV) technology allows road users and the transportation infrastructure to maintain constant communication. This technology is anticipated to increase safety, enhance mobility, and curb the environmental footprint of the transportation sector. However, for these benefits to be realized, data-driven applications need to be developed. This project focuses on using CV data to develop a safety application with a focus on pedestrians–the most vulnerable road users. The goal of this study is to predict a pedestrian’s trajectory so as to intervene and prevent dangerous scenarios that pose safety risk to pedestrians . In order to develop our application, we use data gathered in the form of multi-dimensional trajectories in a variety of contexts. We take into account several kinds of time-series data such as latitude, longitude, velocity, acceleration, rotation etc. This data is then processed to be used as input to a deep learning model that can predict the future trajectory of an agent according to its history trajectory. Deep learning is selected because it allows for predicting future trajectory without making modeling assumptions. To predict the future trajectory, we define a reachable set for each sub-trajectory. Next, we develop a framework called step attention composed of multiple deep-learning models, the output of which is the trajectory of an agent in the next few seconds. The ultimate goal of this application is to alert pedestrians or vehicles of imminent high-risk situations and possibly recommend actions on how to avoid such situations altogether, or reduce the severity of an upcoming incident if it cannot be avoided.