Towards Accurate Vehicle Speed Estimation in Aerial Views: How Localization and Tracking Methods Influence Speed Estimation in UAV-Based Vehicle Monitoring

Abstract

Object detection and tracking workflows are increasingly being explored by researchers in transportation sciences because of their practical applications in the field. These algorithms are widely used with unmanned aerial vehicles (UAVs) to develop robust intelligent traffic monitoring systems. However, benchmarking the speed estimates produced by these cutting-edge technologies has not yet been thoroughly conducted. In this study, we present a critical evaluation of various localization methods namely Axis Aligned Bounding Boxes (AABB) and Oriented Bounding Boxes (OBB) of YOLOv8 paired with different tracking models for vehicle speed estimation. Identifying the optimal detector-tracker combination is essential for the advancement of these autonomous traffic monitoring systems. Through this comparative analysis, the fluctuations in speeds estimated by each detector-tracker combination was evaluated on the basis of the Root Mean Squared Error (RMSE), and Signal Noise Ratio (SNR). The experimental results indicate that trackers combined with OBB-enabled detection exhibited lower signal fluctuations than those based on AABB localization. Furthermore, the selection of the tracker significantly impacts the quality of speed estimates.