A Highly Parallel Hardware Architecture-based Edge Device for Face Recognition and Tracking in Public Places

Abstract

Real-time face recognition has been of great interest in the last decade due to its wide and varied critical applications, which include biometrics, security in public places, and identification in login systems. This has encouraged researchers to deploy recent algorithms on various dedicated hardware devices to develop a fast and accurate face recognition system. However, despite the high efficiency reached in previous works, limitations like the complexity of deep learning models, which result in huge data processing that increases the processing time still exist. To address this problem, this work focuses on implementing a highly parallel and accurate face recognition model by utilizing the hardware resources in nowadays heterogeneous embedded platforms. This constitutes an improvement over previous works, where the tasks are usually allocated to a single engine due to the lack of a unified and automated frameworks that simultaneously explore all hardware engines. The results on real-life video streams suggest that, using simultaneously all the hardware engines that are available in the recent NVIDIA edge Orin GPU and integrating a tracker into the pipeline, a higher throughput of around 290 FPS and a saving of power consumption of around 800 mW have been achieved while satisfying the real-time performance constraint.

Date of Award	2 Jul 2024
Original language	American English
Supervisor	Mahmoud Meribout (Supervisor)