An automated attendance and presence verification system using MTCNN and FaceNet on institutional gatherings

Manual attendance tracking in various gatherings, such as examination halls, seminar rooms, office meeting rooms, conferences, and others, is still widely used in most academic settings due to the lack of adequate technologies that would ensure efficient and reliable presence verification. Current practices include vocal roll calls, attendance sheets, Radio Frequency Identification (RFID), as well as biometric facial and fingerprint scans. Nevertheless, all of these techniques have certain limitations associated with the necessity to interact physically with an individual to confirm attendance or use specific technical equipment, which often leads to inaccuracies and errors when implementing the attendance process. This paper proposes a system called Smart Click that enables automatic attendance control with the help of deep learning (DL)-based face detection and identification algorithms by analyzing a group photo. The system uses the Multi-task Cascaded Convolutional Network to detect several faces on the input image and then processes the detected faces using FaceNet to create 128-dimensional face embedding vectors for each person. FaceNet algorithm identifies people based on the cosine distance comparison to a database containing 128-dimensional vectors of known faces (τ = 0.65). The proposed approach was created and tested on a primary dataset consisting of about 520 photos captured using smartphone cameras from 91 enrolled subjects. Six test cases were conducted to assess performance in different scenarios ranging from face recognition of single subjects to mixed groups and unknown faces. The proposed system managed to achieve an overall recognition accuracy of about 95.26%, False Rejection Rate (FRR) of about 4.76%, and the average processing speed of about 6.34 s per photo on standard hardware without GPU.