Citation

Ahmad, Khubab and Em, Poh Ping and Ab Aziz, Nor Azlina (2025) Harnessing Transfer Learning for Multimodal Driver Drowsiness Detection: CNN-Based Pretrained Models. In: 2025 Multimedia University Engineering Conference, MECON 2025, 21 July 2025 - 23 July 2025, Cyberjaya, Malaysia.

Text 2.pdf - Published Version Restricted to Repository staff only Download (1MB)

Abstract

Driver drowsiness is a significant factor in many road accidents, underscoring the need for accurate and timely detection systems to improve traffic safety. This study proposes a comprehensive framework that integrates On-Board Diagnostic II (OBD-II) sensor data with camera-based imagery to enhance the detection of drowsy driving. Key vehicle parameters, including speed, RPM, throttle position, and steering torque, are synchronised with visual data and transformed into image-like representations using MelFrequency Cepstral Coefficients (MFCC). These representations are processed using pretrained convolutional neural networks: EfficientNetB0, DenseNet201, and ConvNeXtTiny. An additional convolutional layer is introduced at the beginning to convert the MFCC inputs into three-channel images compatible with these models. The models are trained using the Adam optimizer and binary cross-entropy loss function, with support for training enhancements such as checkpointing and adaptive learning rate adjustments. Among the tested architectures, EfficientNetB0 achieved the highest test accuracy of 97.63 percent, followed by DenseNet201 with 97.45 percent and ConvNeXtTiny with 96.81 percent. Evaluation metrics including precision, recall, and F1-score consistently reflect high performance across all models. These results demonstrate the effectiveness of the proposed approach in accurately distinguishing between drowsy and alert driving states. With its high classification accuracy and potential for real-time implementation, this system offers a valuable solution for issuing timely alerts, thereby reducing the risk of fatiguerelated accidents and contributing to safer road environments

Item Type:	Conference or Workshop Item (Paper)
Uncontrolled Keywords:	Convolutional neural network, deep learning,
Subjects:	T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK5101-6720 Telecommunication. Including telegraphy, telephone, radio, radar, television
Divisions:	Faculty of Engineering and Technology (FET)
Depositing User:	Ms Rosnani Abd Wahab
Date Deposited:	18 Mar 2026 08:13
Last Modified:	19 Mar 2026 01:48
URII:	http://shdl.mmu.edu.my/id/eprint/15582

Downloads

Edit (login required)