Reliable Face Recognition on Edge Microcontrollers

This thesis presents the design and implementation of a face recognition system optimized for deployment on edge microcontrollers. The work focuses on achieving a balance between accuracy, speed, and hardware limitations typical of embedded environments. The proposed solution integrates face detection, face recognition, and spoof detection, all running entirely on-device. To meet the strict memory and processing constraints, the models were trained and optimized through quantization techniques and adapted to run efficiently on low-power hardware. The system was specifically developed for the STM32N6 microcontroller, using a combination of public and custom datasets, with particular attention to ensuring realistic conditions and high reliability in face recognition and anti-spoofing tasks. The final prototype is capable of performing all operations locally, displaying the results on an integrated screen. Experimental results demonstrate that the system achieves high recognition accuracy with low latency and energy consumption, making it suitable for real-world applications such as smart access control and embedded security systems.