Synthetic Thermal Image Augmentation and Its Impact on Pedestrian Detection Accuracy

The article presents an algorithm for generating synthetic thermal images with pedestrians and analyses the impact of additionally augmented thermal images on the detection accuracy of pedestrians. A distribution map of annotations was created based on an initial thermal images’ dataset with pedestrians. The study employed the YOLOv11 computer vision model to assess pedestrian detection performance. During the research, the detection accuracy of pedestrians was compared by augmenting the initial thermal pedestrian image dataset with synthetic datasets. The results demonstrate that augmenting the initial thermal image dataset with synthetic data has a positive impact on pedestrian detection accuracy. The additional set of synthesized data resulted in a 1.7% increase in mean average precision (mAP) in our particular research.