Dual-UAV Dataset for Detection and Localization of Armed Individuals

This dataset supports research on dual-UAV human and weapon detection under varying viewpoints and occlusion conditions. Data were collected using two Autel Robotics EVO II Dual 640T V2 UAVs, recording synchronized RGB video at 1920×1080 resolution and 60 FPS. The experimental setup consists of two UAVs observing a single individual wearing a balaclava and holding either a rifle replica (labeled "real") or a visually similar umbrella (labeled "falso"), allowing the evaluation of false positives and visually ambiguous scenarios. During each recording, the subject rotates in 45° increments, generating diverse viewpoints and controlled self-occlusion scenarios. Experiments were conducted under controlled geometric conditions with the following parameters: Target distances: 5 m and 10 mUAV altitudes: 2 m and 5 mInter-UAV viewing angles: 45° and 90° Each filename encodes the experimental configuration using the following pattern: {droneid}_{interuav-angle}_{gt-distance-to-target}_{drone-height} Where droneid identifies the UAV capturing the data, interuav-angle represents the viewing angle between the UAVs, gt-distance-to-target is the ground-truth distance from the UAV to the subject, and drone-height indicates the UAV altitude. Telemetry data and camera information can be extracted from the .MP4 video files via the .SRT subtitle files automatically generated by the UAV system. For convenience, the dataset also provides a .TXT file containing telemetry and camera parameters for the first frame only. This is sufficient for the experiments since the UAVs remained stationary during data acquisition. The file samples.rar contains the exact frame-by-frame image collection used in the experiments reported in the paper "A Dual-UAV Data Fusion Pipeline for Detection and Localization of Armed Individuals from Video". Using this file allows direct reproduction of the experiments described in the paper without preprocessing the original .MP4 video files. To replicate the results, only the telemetry .txt file and the samples images are required. The dataset accompanies the open-source implementation available at:https://github.com/stepsbtw/Dual-UAV-Fusion-Detection-Localization