Automated Attendance Tracking via Multi-Face Recognition and Intelligent Detection

Abstract— This study proposes a hybrid framework for attendance management using multi-face recognition approaches. The proposed method combines a double detection pipeline of Histogram of Oriented Gradients (HOG) and the Haar cascade classifier along with non- maximum suppression (NMS) to improve its detection accuracy with a hybrid FaceNet-based recognition approach that uses deep convolutional neural networks (CNNs) to create robust 128-dimensional embeddings. The final layer of FaceNet is modified to enhance feature separation which improves recognition performance due to the number of people and occlusion typical in real- world situations. The system is capable of detecting and recognizing more than 70 people simultaneously from video streams, static images, or live camera feeds and performs similar real-time processing with detection accuracy and robustness. The unique combination of technical elements allows for system robustness against varying lighting conditions and face occlusion and orientation. To reduce facial dreck to zero the system will employ preprocessing methods that include contrast- limited adaptive histogram equalization (CLAHE) and Gaussian noise reduction. Attendance records are registered into a structured database, and a PDF report is generated automatically and sent through email with one- click to enhance communication for attendance. Our research and system evaluation was performed using a custom dataset constructed from diverse caseform locations for diverse research objectives. The system demonstrated effective high detection accuracy and recognition accuracy, achieving processing speeds of 25ms per frame in real-time. The modular design of the system contributes to its scalability and the intuitive user-friendly interface contributes to the system being an innovative and unprecedented approach towards attendance automation.