Abstract
In mass casualty incidents, the demand for healthcare services far exceeds existing capacity, underscoring the importance of triage. This study supports drone-based triage by developing and validating algorithms for non-contact vital sign measurement from drone footage. Heart rate was measured by extracting average frame colour values from RGB videos and applying signal processing. Respiratory rate was obtained by analyzing temperature changes in the nasal region. Body temperature was analyzed based on the maximum temperature values in the forehead region’ thermal images. For oxygen saturation measurement, a deep learning model trained with features extracted from thermal images. Thirty-seven participants (mean age: 29.7 ± 8.45 years; 64.9% male) were simultaneously recorded using a drone-mounted camera and monitored with a standard bedside reference monitor, with seven recordings obtained outdoors and 30 indoors. The videos were analyzed by creating 15-second segments for respiratory rate and 13-second segments for the other parameters, with 1-second shifts. The accuracy rates of the oxygen saturation, body temperature, heart rate, and respiratory rate measurement algorithms were 98.65%, 98.59%, 97.70%, and 85.22%, respectively, for indoor recordings, and 99.60%, 98.48%, 96.85%, and 82.83% for outdoor recordings. The mean differences between the image processing–based and reference measurements for indoor recordings were 1.26% for oxygen saturation, 0.19 °C for body temperature, -0.37 breaths/min for respiratory rate, and −0.31 beats/min for heart rate; for outdoor recordings, the corresponding mean differences were 0.34%, 0.27 °C, −0.52 breaths/min, and −0.38 beats/min, respectively. The vital sign measurement algorithms demonstrated strong performance and can be successfully integrated into drone-based triage and other remote monitoring systems.