The Computational and Performance Aspects of Masked Face Detection and Recognition

Abstract

This research focuses on developing a mask detection and facial recognition system for real-world applications. The study tests different models for face detection, including Haar Cascade, SSD, HOG, and MTCNN, as well as facial recognition architectures, including VGG, Inception-ResNet-v2, ResNet50, and EfficientNet. The chosen face detection model is the SSD model, which is faster and smaller than other models, and the facial recognition model is FaceNet, a triple lossless model with high accuracy that uses the Inception-ResNet-v2 architecture and includes temperature detection. The final system includes reporting results to the LINE application and was tested with 76 participants from PSU. Wittayanusorn Surat Thani School. The system achieved 99% accuracy in detecting masks and recognized 91% of the faces of all subjects.