Gender recognition using thermal images from UAV

Gender recognition is one of the issues that computer vision deals with. It is useful for analysing human behaviour, intelligent tracking, or human-robot interaction. The aim of this paper is to recognise the gender of people in outdoor areas, where it is very difficult or impossible to guard all access roads to the place, even in poor lighting conditions or in the dark. In this paper, a model will be designed and tested using a controlled UAV flight, during which images of people were obtained. The sensor is a thermal camera located on the UAV, which is not dependent on ambient lighting, and deep learning methods are used for subsequent image processing and classification. These are convolutional neural networks (AlexNet, GoogLeNet), which will be used to solve binary classification. Optimized networks achieve classification accuracy of 81.6 %% (GoogLeNet) and 82.3% (AlexNet). A freely available database [21] was used to learn CNNs, and a self-created database (images obtained with a thermal camera attached to a UAV) was used to test the networks.