FACE RECOGNITION METHODS IN VIDEO SURVEILLANCE SYSTEMS USING MACHINE LEARNING

Abstract

The article is dedicated to the investigation of face identification methods and aims to determine the most suitable one for a security system based on facial recognition from surveillance cameras. The time costs of these methods and their robustness against geometric scale distortions and rotations in various planes have been analyzed. Custom datasets have been generated for experimentation purposes. Special attention has been given to striking a balance between the speed and accuracy of the examined methods for their utilization as the initial stage of a security system based on facial recognition in a video stream. The conducted research has revealed that the most effective methods are RetinaFace-MobileNet0.25, FaceBoxes, SCRFD500MF, and CenterFace; RetinaFaceResNet125, DSFD, and RetinaFaceMobile0.25 which are resilient to facial rotations. Furthermore, when selecting the most optimal facial recognition method for application within a security system, the presence of informative facial parameters was taken into account, as well as the fact that the recognition methods used in the subsequent stage have their limitations concerning resilience to affine transformations.