Deep Learning Based Real Time Face Recognition For University Attendance System

The current face recognition systems struggle with pose, lighting, and expression variations in real-world scenarios. To overcome these challenges, face recognition systems have to apply advanced algorithms and machine learning techniques to analyze and compare facial features. Several studies have proposed face recognition attendance systems based on real-time video processing using deep convolutional neural networks. These systems aim to address problems related to the accuracy rate, stability, and interface settings of the attendance system. In this study, we propose a deep learning approach for face recognition in real-time video data. We constructed a new dataset of real-time videos by automatically and incrementally detecting, labeling, tracing, and purifying faces. We then fine-tune a convolutional neural network with the labeled dataset. Our experimental results on a testing dataset from an attendance management system show that the fine-tuned network achieves an accuracy of 95.3%, outperforming the non-fine-tuned network, which has an accuracy of 81.4%.