ANALYSING ROAD USER TRAJECTORIES AT ROUNDABOUT USING UAV’S

Abstract

A safe and efficient transportation system stands as the cornerstone, propelling the march of a nation towards the zenith of its development. Advances in Unmanned Aerial Vehicle (UAV) technology coupled with the urge of government to examine best practices for improving road safety, require extensive vehicular datasets for a wide range of applications within Intelligent Transportation Systems (ITS). However, the existing datasets partially cover the real world complexities and the representation of wide variety of vehicle types that are present in the traffic streams of developing countries like India. Consequently, the advancement in using aerial imagery for traffic management has been relatively slow. The "Joint Aerial dataset for deTection and trAcking of vehicles in road safety and sUrveillance (JATAYU)" is introduced in this study to address these shortcomings. JATAYU contains almost 1650,000 labeled objects with 55,000 annotated images in ten vehicle classes. The creation of complex algorithms based on massive training and testing datasets is the pinnacle of computer vision technology advancement. The purpose of this study is to compare how well YOLOv8s performs in aerial photography object identification. This research selected VisDrone2019 dataset, VAID and JATAYU dataset.