Real-time person detection from UAV images using performant neural networks

Detection of persons from restricted areas with the help of UAVs is one of the current trends in ensuring the security of regions of interest. For the processing of images acquired by UAVs of special interest are machine learning techniques with an emphasis on neural networks. For this task, two neural networks are proposed and tested in this paper: YOLO v5m and Faster R-CNN. They were learned and tested on two data sets: HERIDAL and a proprietary video stream data set. Both networks gave good results, and the best results were obtained on their own data set (because the images in HERIDAL were taken at a higher height).