A Lightweight HSV Histogram-Based Algorithm for Real-Time Face Recognition on Edge Devices

Abstract: Real-time face recognition systems are increasingly deployed in mobile and edge environments, yet most deep learning approaches demand GPU acceleration and large memory footprints. This paper presents a lightweight algorithm combining Haar Cascade detection, HSV histogram feature extraction, and cosine similarity classification. Implemented via a Flutter mobile client and Flask REST API backend, the system achieves an average end-to-end latency of 82.5 ms and a False Acceptance Rate (FAR) of 0.25% without GPU support. Experimental evaluation demonstrates that algorithmic efficiency and minimal infrastructure overhead can outweigh marginal accuracy gains of deep learning models in controlled access-control scenarios.