DBEV-UAV: Dual Bird's-Eye View RGB\u2013Depth Dataset for UAV Detection and Tracking

Real-time UAV detection using onboard 360\u00b0 LiDAR sensors faces a critical computational bottleneck: processing raw point cloud data for drone detection requires high computational resources, resulting in insufficient frame rates for real-time tracking applications. We present DBEV-UAV, a dual-modality dataset that addresses this challenge by transforming 3D LiDAR point clouds into 2D bird's-eye view (BEV) depth maps, enabling efficient YOLO-based detection with significantly improved inference speeds.The dataset comprises two modalities: (1) simulated BEV depth maps generated from 360\u00b0 LiDAR point clouds in a controlled virtual environment, and (2) real-world RGB images of actual UAVs. The depth map modality enables real-time detection across the full 360\u00b0 field-of-view by converting computationally expensive 3D point cloud processing into efficient 2D image detection using YOLO. While depth map detection achieves high-speed inference, the RGB modality provides higher accuracy detection within the camera's limited field-of-view, creating a complementary dual-sensor detection system.This dual-modality approach solves a critical operational gap in UAV pursuit scenarios: when a target UAV maneuvers outside the RGB camera's field-of-view, the BEV depth map maintains continuous detection coverage across the full 360\u00b0 range until the pursuing UAV reorients itself. The trained YOLO model achieves real-time detection performance on depth maps while RGB detection ensures accuracy within overlapping coverage areas. DBEV-UAV provides researchers with a benchmark dataset for developing efficient multi-modal UAV detection algorithms that balance computational efficiency, spatial coverage, and detection accuracy for autonomous pursuit and tracking applications.