DRaT: Drone and Radar Trajectories - A Dual-Modality Dataset for Highway Traffic Monitoring

DRaT (Drone and Radar Trajectories) is a dual-modality dataset of naturalistic vehicle trajectories designed to systematically evaluate infrastructure-based radar sensing against drone-derived ground truth. The data were collected at a highway merging segment along the Ronald Reagan Memorial Highway in Fort Worth, Texas, covering a roughly 140 m (460 ft) eastbound section that transitions from four lanes to three. Collection summary - Duration: ~80 minutes of synchronized collection — a 43-minute free-flow period (density ≈ 40.8 veh/km, space-mean speed ≈ 95.3 km/h) and a 37-minute congested period (density ≈ 122.9 veh/km, space-mean speed ≈ 28.7 km/h).- Drone (ground truth): hovered at 122 m (400 ft) AGL, capturing 4K video at 30 fps from a bird's-eye view. Trajectories were extracted using the open-source geotrax framework (https://github.com/rfonod/geo-trax), which combines a YOLO detector with the BoT-SORT multi-object tracker; frames were stabilized to a master reference and georeferenced to an orthophoto via ground control points.- Radar: a commercial FMCW K-band radar (24.25–24.50 GHz), 110° horizontal field-of-view, mounted at ~5.5 m height with a 15° downward tilt. Outputs object-level positions, speeds, dimensions, and headings at 8 Hz, georeferenced using an RTK-corrected GPS reference and a refined azimuth angle.- Temporal alignment: drone trajectories (30 Hz) were linearly interpolated onto the radar's 8 Hz timestamps for one-to-one correspondence. Contents - drone_GT.csv — 5,278 drone-derived ground-truth trajectories (392,707 detections aligned to radar timestamps).- radar.csv — 5,430 radar-derived trajectories (282,712 detections at 8 Hz). Both files share a common core schema: vehicle ID, timestamp (Unix epoch, 8 Hz), WGS84 lat/lon, UTM zone 14N coordinates, the radar's sensor-centered local frame in feet, scalar speed in m/s, and vehicle dimensions in meters. The drone file additionally provides acceleration and a detector-assigned vehicle class. The radar file additionally provides a radar object-type code, per-track quality, and a smoothed lane assignment. Intended use - Multi-level evaluation of radar detection and tracking (individual vehicle, trajectory, macroscopic flow).- Trajectory-level research based on infrastructure sensing systems, particularly in merging areas where lane-changing and dense maneuvers are common.