Applying Face Recognition in Video Surveillance for Security Systems

Abstract

In order to meet the security needs that are becoming more and more important with the economic advances, the development of physical or biometric access control systems is constantly growing. Several biometric modalities can be used and each one presents a particular interest, according to the targeted application. Within the framework of our study paper, we have realized a facial recognition system based on the EfficientDet model following the architecture of a deep neural network. The facial recognition process is divided into several steps, namely: face detection in each image, face normalization, facial feature extraction, classification and decision. The training and evaluation of the system were done on the database: Casia-web face. As Casia-web Face is unlabelled, we have developed an algorithm using the open source deep learning framework Mxnet to convert the images into binary format, reduce their size and give each image an identifier. Finally, the optimization of the system has been done using Root Mean Squared Propagation (RMSProp) and the Shard shuffling optimizers.