TinyML-Based Facial Recognition for Embedded Systems

Facial recognition is widely used in security and biometric applications but remains challenging to deploy on resource-constrained embedded systems. This paper presents a TinyML-based facial recognition system optimized for lowpower microcontrollers using MobileNetV2, EfficientNet, and a simplified CNN. The proposed system applies pruning and quantization to reduce model size and computational load. Among the evaluated models, the simplified CNN achieved the best trade-off, maintaining 97.9 % accuracy after compression with only 145 KB memory usage, making it ideal for deployment on the Arduino Nano 33 BLE Sense. These results highlight the potential of TinyML for enabling real-time, efficient facial recognition on edge devices, despite limitations in hardware and environmental conditions.