Evaluation of Face Recognition Schemes for Low-computation IoT System Design

Abstract

The authenticated door-lock security system for smart-home is a critical Internet-of-Things (IoT) application. The most pressing bottleneck of running algorithms in IoT devices is the availability of low computation resources. The state-of-the-art algorithms which work significantly well in high-end computing unit more often fail to operate in IoT devices like Raspberry pi. This work targets design of a face authenticated door-lock security system in Raspberry pi and, therefore, evaluates the performance of conventional face detection and recognition models. The local binary pattern histogram (LBPH), eigenface and fisherface algorithms are considered for evaluation. The evaluation analyzes the accuracy of a model in detecting human faces, its computation time and vulnerability to photo hacking while implemented in Raspberry pi platform. It covers a large number of stored faces and the cases of variation in angle and illumination in the face image. The python and c++ bindings to the opencv framework are considered to create a fully functional face recognition system hardware.