MobileFaceNet-Based Facial Recognition System for Contactless Access Control

Abstract

Conventional access control systems require physical contact, such as pressing buttons, using fingerprints, or employing radio-frequency identification (RFID). These interactions increase the risk of transmitting diseases like COVID-19 through surface contact. To mitigate this risk, a contactless identification method becomes imperative. Hence, this study proposes an adaptive and seamless dual-mode mobile facial recognition-based access control system. Specifically, the proposed system uses a convolutional neural network (CNN) algorithm (MobileFaceNet) to develop high-precision facial verification on mobile devices with a dual-mode (real-time and offline) facial recognition system. The system is implemented on Raspberry Pi, running an Android Operating System equipped with camera functionality. The system instantaneously operates by detecting human faces, compares the faces with a pre-existing database, and grants access to successful matches. The system underwent preliminary testing on an Android 9 device with an 8-megapixel camera. Faces were detected and recognised in a record time (within seconds), showcasing commendable accuracy. The system achieved a recognition accuracy of 95% under varied lighting conditions and distances. The proposed system can be seamlessly integrated with electronic doors, ensuring optimal access control for smart homes and restricted facilities.