<![CDATA[Manually identifying graduates can slow down the procession and increase the risk of misidentification. This study evaluates a real-time facial recognition system using VGG-Face for feature extraction and Cosine Distance for matching. Faces are detected using OpenCV's Haar Cascade and represented as 4096-dimensional embeddings via the VGG-Face model (accessed through DeepFace). Identity is determined based on the minimum Cosine Distance value using a 0.6 threshold and validated by consistent detection across three consecutive frames. Testing involved 20 individuals (10 registered graduates and 10 unregistered individuals) across four conditions: distances of 50 cm, 100 cm, and 100–150 cm, as well as a 50 cm distance with the subject wearing glasses. Out of 885 recorded recognition events, the results yielded 416 True Positive s, 440 True Negatives, 22 false Positive s, and 7 false Negatives. Evaluation metrics showed an accuracy of 96.7232%, precision of 94.9772%, recall of 98.3452%, and an F1-score of 96.6318%. The average system latency was 0.6877 seconds. For the registered group, varying the distance up to 150 cm resulted in minimal changes to the recognition rate, whereas wearing glasses reduced the recognition rate to 95%. The findings indicate that the system supports rapid graduate identification in semi-controlled environments, however, the rejection of unregistered faces and robustness against occlusion require further improvement.]]>
Copyrights © 2026
