This work presents the implementation of a face verification system designed for edge hardware, where mmWave radar is used to trigger image capture and a compressed deep learning model performs local recognition. The system integrates a 24 GHz mmWave sensor for intelligent human presence detection, an ESP32-CAM module for image capture and preprocessing, and an ESP32-S3 microcontroller for real-time face embedding extraction and matching. A knowledge distillation framework compresses a MobileFaceNet 1.0× model into lightweight student variants enabling INT8 quantized inference fully on-device. Experimental results demonstrate sub-second recognition delay with minimal accuracy degradation, making the system suitable for smart home and IoT-based access control applications.
N et al. (Thu,) studied this question.