Abstract
Manual attendance registers in educational systems is inefficient and prone to human error; therefore, they need an automated solution that is efficient and accurate. This paper describes a hybrid face recognition system combining FaceNets deep convolution neural network with triplet loss, which extracts 128-dimensional face embeddings. The attendance monitoring system has a dual input capability of group photo uploads where 2-3 images are combined into a single file to improve face detection accuracy, and live attendance captured via specific camera utilization in a web application. The face detection part of the program uses Open CVs Haar Cascade and the attendance is stored as embedding's in a SQLite database for fast face matching. The report offers an online dashboard with real time attendance reporting, historical attendance records, and CSV file download capabilities. Implemented on a dataset of 30 images of students followed by (5) live feeds, the system achieved 98.5% accuracy which we believe is better than current manual and automatic methods date-to-date (HOG+PCA of 85.7%). The proposed method has improved scalability, improves potential security, and provides ease of administration in educational settings.