Attendance recording and verification are essential in educational institutions, especially during examinations, where ensuring each student’s presence is important. Conventional attendance systems rely on manual methods such as name calls or signature verification, which are often time-consuming and prone to errors. Although various facial recognition systems have been introduced to address these challenges, many state-of-the-art models still struggle to accurately identify students under varying conditions, including differences in lighting, head orientations, and facial appearances influenced by accessories. To overcome these limitations, this study proposes a multi-facet facial recognition model based on the K-Nearest Neighbors (KNN) algorithm for recording and verifying student attendance during examination scenarios. The model was trained to operate reliably under different lighting, head orientations, and use of accessories, specifically glasses and beards, by leveraging pre-registered students’ image data captured across multiple settings, forming a training dataset of 990 images. The process involved face detection using a pre-trained Haar Cascade classifier, followed by face alignment and encoding into 128-dimensional vectors using a ResNet-based approach, which were then classified by the KNN algorithm. Evaluation across static images, video, and real-time camera feeds achieved an overall accuracy of 99.00% with a response time of 6.00s seconds. Compared to traditional manual attendance systems, the proposed model provides a solution for examination verification, effectively mitigating recognition challenges caused by varying environmental and facial conditions.
Jack et al. (Sun,) studied this question.