A low-computational cost face recognition system for high accuracy privileged access management to support the development of smart laboratories
Muhamad Yusvin Mustar, Karisma Trinanda Putra, Yudhi Ardiyanto, Rama Okta Wiyagi, Ibnu Arseno, Hsing‐Chung Chen · AIP conference proceedings · 2023
Nowadays, privileged access management (PAM) is important for the development of smart laboratories, especially in modern universities and research centers.A biometrics-based PAM such as a face identification system is needed to safeguard identities to define specialized capabilities beyond regular users while accessing the labs or the equipment.However, real-world PAM implementation faces high computational costs and a rigid centralized system, making its scalability and the depth of implementation limited.This study proposes a design of a PAM system using face identification technology and low-powered embedded systems.A prototype is made using Raspberry Pi 4 as a computing module and a webcam as a visual sensor.The design considers scalability and performance through an edge-computing scheme and a low-complexity algorithm.The design is developed through two-stage architecture i.e., face detection and face recognition.The face detection algorithm is designed through Haar cascade classifier (HCC).In addition, the face recognition algorithm is performed using local binary pattern histogram (LBPH).An analysis is provided by comparing the methodology with deep learning (DL)-based algorithm.The average accuracy using 5-fold cross-validation using LBPH is 3 -10 % lower than those using the DL model.Although the accuracy using LBPH is slightly lower, in real-time applications, the prototype can recognize lab member faces at a maximum distance of one meter with a frame rate reaches 10 FPS i.e., higher than those achieved by the DL model.Furthermore, the system can be enhanced with a centralized supervisory system that connects to the edge computing node making its scalability more manageable.