Temporal Attention-Based Multimodal Networks for Real-Time Proctoring of Online Examinations

Abstract

The abstract should summarise the contents of the paper in short terms, i.e., 150-250 words. The growth of online education has driven the need for technologies that uphold academic integrity in remote online assessments. While traditional online proctoring methods may not be the most resource-efficient, this paper presents a multimodal neural network-based online proctoring system that processes webcam recordings to detect suspicious device usage. The neural network is trained on a dataset of 18 subjects, with videos manually annotated for cheating and non-cheating behaviours. Various features are extracted, including head pose via 6DRepNet, eye gaze using L2CS-Net, facial landmarks through the MediaPipe library, and cropped faces detected using YOLO-v11. Subject embeddings are generated using ResNet50 and reduced with PCA to minimise computational complexity. The proposed multimodal neural network achieves an overall classification accuracy of 85.74%, with late-fusion accuracy across system components reaching 94%.

DOI: https://doi.org/10.24321/3117.4841.202602