Collective Perception with Unmanned Aerial Vehicles for Smart Intersections

Collective Perception can increase the field of view of automated vehicles. For busy intersections, infrastructure-based perception and communication can improve the safety. While building Roadside Units for Collective Perception is costly, Unmanned Aerial Vehicles (UAVs) can support road users with higher flexibility. This paper presents a method for aerial object detection and V2X communication with 5G. The focus is on UAVs as additional sensor platforms to detect traffic objects and obstacles and share this information via Collective Perception Messages (CPMs). A prototypical implementation that combines detection and communication in a controlled environment is realized. It includes the development and training of an object detection model for the specific perspective of a UAV and the integration of the recognized objects into CPMs for the transmission with 5 G communication. The trained models are tested on videos with UAV footage for evaluation of the generalization capability on new cameras. Our results show that a lightweight YOLO model trained for vehicle object detection in the traffic system provides high precision to enable Collective Perception.