Field experiment dataset for vision-assisted radar tracking of non-cooperative aerial vehicles: tracking evaluation, YOLO detector component, and camera-radar calibration

Research on unmanned aerial vehicle perception for airborne deployment, using 4D millimeter wave radar and monocular camera to track non cooperative small unmanned aerial vehicles in the field. RTK-GNSS is only used as an offline evaluation reference.

The dataset includes: (1) frame by frame evaluation results of radar only and visual assistance schemes on 14 external field sequences (5 close range dual target D series, 4 close range single target S series, and 5 far range single target L series), including output purity and coverage under a 5-meter threshold F1、 False confirmation output per frame, matching sample 3D RMSE); (2) Comparison results of six radar measurement construction and allocation strategies on 17854 common evaluation frames in the D00-D04 binocular target subset (including source point duplicate consumption count); (3) Assembly ablation and distance segmentation error statistics for projection margin gate and trajectory confirmation; (4) YOLOv8n UAV Freeze Detector Component (model weights, 232 independent test images and 280 annotated instances, standard evaluation report, detection accuracy mAP50 0.929 can be end-to-end recalculated); (5) Camera radar joint calibration data (checkerboard calibration image, solver audit summary, reprojection verification map, calibration experiment report); (6) Complete audit chain (candidate graph hash verification, running list, audit report, checksum), all numbers in the paper can be recalculated by the script within the package.