A RELIABLE MASKED FACE RECOGNITION SYSTEM FOR PERSON’S SAFETY AT PUBLIC PLACES

Abstract

Wearing face masks in public places has become common since the COVID-19 pandemic. It provides the safety of a person from various viral diseases and dust in polluted spaces. Wearing face masks in public places is a significant concern for public safety as it has become a pervasive pattern that a person comes wearing a mask, performs a crime, and escapes. This crime pattern is a significant concern for security agencies for surveillance work, identity verification, and many other tasks. Identifying that person has become a tedious task for the security agency. Therefore, to identify a person even while they are wearing a face mask, an automated and efficient masked facial recognition system is required. This research proposed a reliable model to identify a person’s identity even when wearing a facemask using the FaceNet model. The proposed model in this work enhances security by incorporating a robust pipeline that includes masked face detection, alignment, and embedding extraction using a pre-trained FaceNet network. These embeddings have been compared using distance-based metrics to verify or identify individuals, even when facial regions are partially occluded. The effectiveness of the suggested work is confirmed by testing using masked images, which yields an accuracy of over 97 % when the model is trained using either masked images or a combination of masked and unmasked images. Results from experiments demonstrate that even when a person is wearing a facemask, the suggested model can correctly identify them.