Face Localization and Recognition on DronesUsing Deep Learning

Unmanned aerial vehicles (UAVs) search dangerous or difficult places for humans and collect various environmental data. Human face recognition in drones is essential for various applications, such as surveillance, search, and security. Previous methods for face recognition are highly sensitive to limitations such as height, angle and distance from the face. Therefore, locating and identifying faces at high altitudes and long distances reduces the accuracy and efficiency of previous methods. In this paper, a new deep leanring-based face detection and recognition is presented. The proposed method is performed in three steps. In the first step, input images are segmented with the selective search algorithm. In the second step, a deep network is proposed for filtering bounding boxs to identify the target boxes with high accuracy and speed. In fact, a two-class classification problem is performed by deep learning forfacelocalization. In the third step, the localized faces are usedto perform face recognition using the proposed deep network. In the proposed architecture, the properties of widely used deep networks are used, and a quantitative comparison of the proposed method with new methods in terms of computational complexity shows that training the proposed model requires less runing time than other methods. In addition, the evaluation of the proposed method on the DroneFace dataset for different distance and height from the target shows thatthe proposed method has an average face detection rate of 75.9 and an average face recognition rate of 84.6. Therefore, the proposed method has higher accuracy and efficiency than state-of-the-art methods in this field and can be used for surveillance and security applications.