| Penulis/Author |
Gheri Febri Ananda (1); Prof. Ir. Hanung Adi Nugroho, S.T., M.Eng., Ph.D., IPM., SMIEEE. (2); Dr.Eng. Ir. Igi Ardiyanto, S.T., M.Eng. (3) |
| Abstrak/Abstract |
Face detection in educational environments is
vital for applications such as automated attendance systems,
behavior analysis, and personalized learning tools. This study
conducted a comprehensive comparative analysis of four
advanced face detection algorithms—Haar Cascade, MTCNN,
YOLOFace, and RetinaFace—using a dataset of real classroom
images. The analysis addresses the unique challenges of
classroom settings, including variable lighting, occlusions, and
diverse student orientations. Metrics such as precision, recall,
F1-score, and inference time were evaluated. YOLOFace and
RetinaFace demonstrated superior performance, with
YOLOFace achieving perfect precision, recall, and F1-score in
most scenarios. RetinaFace also demonstrated high precision
and recall, though its inference times were longer, up to 12.73
seconds. MTCNN exhibited high precision, up to 1.00, and
recall, up to 0.97, but had higher inference times and
occasionally missed some faces. Haar Cascade, while efficient
with the shortest inference times, down to 0.54 seconds,
displayed lower precision of 0.89 and recall of 0.25, resulting in
a higher rate of missed detections. These findings emphasize the
importance of selecting appropriate face detection models that
address the specific challenges of educational environments to
enhance the effectiveness of digital classroom applications. |
