Evaluating the Accuracy of Vehicle Tracking Data from Unmanned Aerial Vehicle Videos and Identifying Traffic Conflict Points to Develop a Conflict Frequency Predictive Model for Roundabout Safety Assessment under Mixed Traffic

Abstract

In recent years, low-and-medium-income countries have been experiencing widespread use of unmanned aerial vehicles (UAVs) in the field of traffic engineering and management studies. However, the highly heterogeneous traffic mix poses challenges for UAV-based automated analyses. This study stands out by evaluating the accuracy of vehicle tracking data collected using UAVs for traffic safety assessment. High-resolution aerial imagery and precise ground truth data were obtained using a DJI Mavic Mini 3 Pro drone and Racelogic VBOX 3i Global Positioning System device. The YOLOv12x model was used for vehicle detection, while ByteTrack ensured continuous tracking across frames. Speed data extracted from UAV footage were validated against ground truth measurements and trajectory data from DataFromSky, showing low mean square error, root mean square error, normalized root mean square error, and mean absolute percentage error values. Conflict points at roundabouts were identified using surrogate safety measures such as modified time to collision and deceleration rate to avoid collision. The spatial distribution of the conflict points was visualized through a heatmap. A conflict frequency prediction model was developed using an AdaBoost algorithm with a negative binomial objective function, achieving high predictive accuracy. Feature importance analysis, along with the SHAP plot, identified the most critical geometric and traffic variables influencing conflict frequency. This work is unique in its application of advanced UAV technology combined with cutting-edge detection and tracking algorithms for comprehensive traffic safety analysis. By demonstrating the potential of UAVs in accurately monitoring traffic dynamics and identifying conflict points and conflict contributory factors, this study paves the way for innovative approaches to enhance road safety.