Application of Data Augmentation Methods to Unmanned Aerial Vehicle Monitoring System for Facial Camouflage Recognition

Recently, the Unmanned Aerial Vehicle (UAV) monitoring system based on face recognition technology has attracted much attention. However, partly because of human hair changes, glasses wearing and other camouflage behavior, the accuracy of UAV face recognition system is still not high enough. In this paper, two kinds of data augmentation methods (the hairstyle hypothesis and eyeglass hypothesis) are used to expand the face dataset to make up the shortage of the original face data. In addition, the UAV locates human's face in the air from special distance and elevation, the collected face characteristics are vastly different from those in the public face library. Considering the peculiarity of UAV face localization, the data augmentation program is implemented to improve the accuracy of UAV identification of camouflage face to be 97.5%. The results show that our approach is effective and feasible.