Abstract:

In this paper, the task of automatic person detection in thermal images using convolutional neural network-based models originally intended for detection in RGB images is investigated. The performance of the standard YOLOv3 model is compared with custom trained model on a dataset of thermal images extracted from videos recorded at night in clear weather, rain, and fog, at different ranges and with different types of movement – running, walking and sneaking. The experiments show excellent results in terms of average precision for all tested scenarios, and a significant improvement of performance for person detection in thermal imaging with a modest training set.