Innovative AI system analyzes collision data from traffic-cameras

DOWNTOWN BROOKLYN — TRAFFIC COLLISIONS AND NEAR-MISSES COULD BE MORE CLOSELY MONITORED, with a new AI monitoring system that NYU Tandon researchers have developed.

The system, named SeeUnsafe, scans existing traffic-camera footage to detect not only collisions but also near-misses. The SeeUnsafe program combines language reasoning and visual analysis, providing transportation agencies a low-cost way to identify dangerous intersections. The SeeUnsafe system has the potential to assist cities in improving road safety without having to make major financial investments.

Kaan Ozbay, director of NYU Tandon’s C2SMART center and the paper’s senior author, presented this study at the eighth annual Research on the Road symposium on Nov. 19. The project won New York City’s Vision Zero Research Award, an annual recognition of work within the framework of the city’s road safety priorities. The paper’s co-author is NYU Tandon Associate Professor Chen Feng, a co-founding director of the Center for Robotics and Embodied Intelligence.

SeeUnsafe outperformed other AI models on a Toyota Woven Traffic Safety dataset test, in which the program accurately classified videos as collisions, near-misses, or normal traffic 76.71% of the time.

The National Science Foundation and the U.S. Department of Transportation’s University Transportation Centers Program provided funding for the research.

✰✰✰