Face Recognition on Edge Devices

Abstract

In the rapidly evolving landscape of face recognition technologies, this project addresses the pressing need to evaluate and optimize diverse face recognition models for deployment on edge devices. Spanning traditional methods such as Eigenfaces to contemporary deep learning approaches like VGG Face, DeepFace, and ArcFace, the study conducts a rigorous comparative analysis across various operational settings. The central focus is on elucidating the impact of Graphics Processing Units (GPUs) in enhancing model performance, particularly within the resource constraints inherent to edge devices. Empirical testing unfolds on a dataset that’s widely recognized as the Labeled Faces in the Wild (LFW) dataset. Employing Python as the primary programming language, TensorFlow as the machine learning backbone, and leveraging the capabilities of Keras, OpenCV, and Adam Geitgey’s Face Recognition library, the project assembles a versatile toolkit for model evaluation. Upon completion, this project is poised to contribute substantial insights into the deployability of face recognition models on edge devices, providing practical guidance for developers, engineers, and decision-makers. The anticipated outcomes encompass a nuanced understanding of the models’ adaptability, limitations, and the role of GPUs in enhancing performance. As technology converges towards decentralized computing paradigms, this study is positioned to play a pivotal role in shaping the landscape of face recognition technologies on the edge.