Airborne Multi-Sensor Dataset for Medical Emergency Response: A Resource for Computer Vision Research

This work introduces a novel dataset to advance computer vision applications for emergency medical response using Uncrewed Aircraft Systems (UAS). The dataset addresses the critical need for real-world aerial imagery of simulated medical emergency scenarios, collected via a helicopter platform equipped with electro-optical (EO), infrared (IR), and Inertial Navigation System Global Positioning System (INS-GPS) sensors. Fifteen distinct scenarios were captured, featuring diverse arrangements of people in varying poses (standing, sitting, supine, prone, signaling, kneeling, and walking) across two environments: a grassy area and a beach. The data acquisition process involved controlled helicopter approaches and orbits using an externally mounted sensor payload to capture comprehensive scene information. This paper details the flight campaign, sensor suite, data collected, and annotation process. Overall, approximately 284,724 frames of video data were collected directly related to human stance recognition across three camera sensors, with 10,000 frames annotated with bounding boxes classifying the stances of the visible human subjects.