UAVDB: Trajectory-Guided Adaptable Bounding Boxes for UAV Detection

The widespread deployment of Unmanned Aerial Vehicles (UAVs) in surveillance,security, and airspace management has created an urgent demand for precise,scalable, and efficient UAV detection. However, existing datasets often sufferfrom limited scale diversity and inaccurate annotations, hindering robust modeldevelopment. This paper introduces UAVDB, a high-resolution UAV detectiondataset constructed using Patch Intensity Convergence (PIC). This noveltechnique automatically generates high-fidelity bounding box annotations fromUAV trajectory data~\cite{li2020reconstruction}, eliminating the need formanual labeling. UAVDB features single-class annotations with a fixed-camerasetup and consists of RGB frames capturing UAVs across various scales, fromlarge-scale UAVs to near-single-pixel representations, along with challengingbackgrounds that pose difficulties for modern detectors. We first validate theaccuracy and efficiency of PIC-generated bounding boxes by comparingIntersection over Union (IoU) performance and runtime against alternativeannotation methods, demonstrating that PIC achieves higher annotation accuracywhile being more efficient. Subsequently, we benchmark UAVDB usingstate-of-the-art (SOTA) YOLO-series detectors, establishing UAVDB as a valuableresource for advancing long-range and high-resolution UAV detection.