Multimodal Biometric Attendance System Using Convolutional Neural Networks with Face and Fingerprint Fusion

Manual roll-call and single-trait attendance systems such as RFID cards or lone fingerprint scanners are slow and can be defeated through proxy attendance, in which one person marks attendance for an absent friend. This paper presents a Multimodal Biometric Attendance System that combines live face recognition with fingerprint recognition using Convolutional Neural Networks (CNN). Two CNN branches, one for face images and one for fingerprint images, are trained separately and then joined by a feature-level fusion model that makes a single combined identity decision. Face images were captured from team members through a webcam, and fingerprint samples were taken from the public SOCOFing dataset because no physical scanner was available. All images are converted to 100 × 100 grayscale and expanded with OpenCV-based data augmentation to reduce overfitting. The system is deployed as a Flask web application with a live camera interface, an SQLite attendance database, proof-photo capture, and a password-protected administrator dashboard with search, export, editing, analytics, and low-attendance alerts. A face-first decision strategy with a confidence and margin check was introduced after early versions misidentified users. The best fusion model reached about 99% validation accuracy. The framework provides a practical, end-to-end and extensible approach to secure automated attendance that can later accept a real fingerprint scanner.