Potential of infrastructure-based sensors to road safety

Abstract

Recent developments in sensor and computer technology enable roadside units not only to monitor traffic flow, but also to process the collected data to improve road safety. This paper presents an implemented test field with a structure of its use cases ranging from traffic monitoring to advanced extended perception utilizing a vehicle to infrastructure communication techniques. The traffic monitoring use case can be extended with an automatic accident detection being performed in a road-side unit. The accident detection method is to identify a potential accident in near real time in order to activate the first responders without unnecessary delay. The proposed vehicle-to-vehicle accident detection method is based on the evaluation of sudden velocity vector differences. The vehicles are recognized using the YOLO method from the video stream and the optical flow method is used to evaluate the sudden velocity changes, which are classified in order to decide about a possible accident.