Face recognition system based on a single-board computer

With state-of-the-art computing systems along with Graphical Processing Units a Deep Neural Network can be realized by training on any publicly available dataset in order to detect faces. In a real-time application, the inference of such a neural network may not require high computational power as while in the training procedure. In this paper, the authors proposed face recognition and face detection system based on a single-board computer. Several different single-board computers like Raspberry Pi, Banana Pi and Nvidia Jetson Nano were evaluated. The authors compared two different face detection algorithms. These two algorithms are Haar feature-based cascade classifier and the second one is a multitask cascaded convolutional neural network (MTCNN). As a face recognition algorithm, the authors used FaceNet, which directly learns a mapping from face images to a compact Euclidean space where distances correspond to a measure of face similarity. FaceNet outputs embeddings as feature vectors which are feed to face classification algorithm. The system is trained and tested on a proprietary database and is installed to monitor who enters the room. The system has over 97% accuracy. The goal of this paper is to convey the possibility of successfully incorporating face recognition and face detection systems in small, low-power devices.