This paper presents a comprehensive study of human detection using thermal video data captured from Unmanned Aerial Vehicles (UAVs), usually drones. The data collection was performed during the daytime. The methodology involved the acquisition of drone-generated videos, frame extraction, annotation with bounding boxes, and dataset augmentation to enhance the robustness of the model. A custom dataset was created using the Autel Evo II Enterprise drone with a built-in thermal imaging camera. The experiments utilized You Look Only Once (YOLO) pre-trained models, specifically YOLO5s and YOLO8m, with a detailed analysis of precision, recall, and mean Average Precision (mAP). The findings suggest that YOLO8m outperforms YOLO5s, showcasing superior results in terms of precision, recall, mAP(50), and mAP(50–95), while effectively minimizing the false positive rate.