Stop-and-Go Trajectory Data of Disordered Traffic Using a Swarm of Unmanned Aerial Vehicles

Studying traffic dynamics and behaviour requires detailed information on vehicular trajectories. However, vehicle trajectory extraction is challenging in congested and disordered traffic due to false detections and tracking errors caused by closely spaced platoons and vehicle occlusion. Unmanned aerial vehicles (UAV) give a bird-eye view of traffic from which vehicular trajectories can be effectively extracted. The most studies employed a single UAV to capture data, limiting the trajectory information to small road segments, which may not reflect the complete traffic dynamics and naturalistic driving behaviour. This study aims to collect stop-and-go traffic data in disordered traffic conditions over a longer road section using a swarm of UAVs. The data were collected on a six-lane divided urban arterial road in Chennai city, India. Furthermore, the study presents a process for collecting vehicle trajectory data from video feeds of a swarm of UAVs. A pre-processing method based on feature transformation for unifying frames from individual UAVs covering a long road section has also been developed. A framework for effectively detecting vehicles in large pixel frames at congested states has been proposed using a supervised learning-based method and the Slicing Aided Hyper Inference algorithm in a pipeline. An observation-centric tracking algorithm was employed to mitigate the tracking errors. Detailed performance evaluation of the presented framework has been done on the collected dataset. The results show effective detection and tracking of vehicles in congested and disordered traffic, giving high-precision vehicle trajectories.