Multiperson Automated Attendance System Based on Face Recognition Using YOLO and DeepFace with Active Learning

Suprihanto, Didit
Wirawan, Adi Pandu
Saputra, Kahlil Gibran
Harjanto, Arif
Hakim, Imam Muhammad
Nugroho, Happy

International Journal of Engineering, Science and Information Technology
Vol 5, No 4 (2025)

Accurate student attendance tracking is essential in academic environments, yet traditional methods remain inefficient and vulnerable to manipulation. This research presents a classroom attendance system based on facial recognition that integrates YOLO for multiperson face detection and the SFace model from the DeepFace framework for feature extraction and identity matching. A key contribution of this study is the implementation of an Active Learning mechanism that enables the system to update its embedding Database using user-provided corrections, enabling continuous adaptation to real classroom conditions. The system was developed as a Python-based desktop application and evaluated using 38 group images captured with various devices under uncontrolled lighting, diverse head poses, occlusion, and different classroom densities. Performance was assessed using accuracy, False Rejection Rate (FRR), and False Acceptance Rate (FAR) across two scenarios: before and after Active Learning. Experimental results show a substantial improvement after the learning process, with accuracy increasing from 52.0% to 96.6%, while maintaining a low FAR of 0%. These findings demonstrate that Active Learning effectively enhances recognition performance by enriching the embedding Database with real-world facial variations that do not present during initial registration. Overall, the proposed system highlights the importance of integrating Active Learning into face recognition–based attendance applications to improve robustness and adaptability in unconstrained multiperson classroom environments.