Abstract: Traditional attendance management methods in academic and organizational settings are labour-intensive, error-prone, and vulnerable to proxy attendance. This paper presents a Smart Attendance System (SAS) that leverages deep learning-based facial recognition to automate and secure the attendance process. The proposed system employs a Residual Convolutional Network (RCN) for feature extraction, combined with FaceNet embeddings and a Support Vector Machine (SVM) classifier for identity verification. Real-time face detection is achieved using Haar Cascade and Multi-task Cascaded Convolutional Networks (MTCNN), while OpenCV handles video frame acquisition. The system is engineered to capture attendance within a bounded temporal window (25–30 minutes) that corresponds to a single class session, thereby eliminating duplicate entries. Experimental evaluation on a dataset of 200 subjects yields a mean recognition accuracy of 97.4%, a false acceptance rate of 0.8%, and an average processing latency of 1.2 seconds per frame. The system stores records in a relational database and provides administrators with export and reporting capabilities. Results demonstrate that the proposed architecture outperforms conventional LBPH and Eigenface baselines by a statistically significant margin, offering a scalable, contactless, and cost-effective solution for modern smart institutions.

Keywords: facial recognition; deep learning; convolutional neural network; FaceNet; attendance automation; MTCNN; OpenCV; residual network; smart classroom; biometric system.

Downloads: | DOI: 10.17148/IJIREEICE.2026.14414