EffRes-DrowsyNet: A Novel Hybrid Deep Learning Model Combining EfficientNetB0 and ResNet50 for Driver Drowsiness...

Area of Science:

• Computer Vision
• Artificial Intelligence
• Machine Learning

Background:

• Driver drowsiness is a significant cause of road accidents, leading to delayed reactions and impaired cognitive function.
• Current driver monitoring systems require improvement for real-time, accurate fatigue detection.

Purpose of the Study:

• To introduce EffRes-DrowsyNet, a novel hybrid deep learning model for early driver drowsiness detection.
• To evaluate the model's performance across diverse datasets and conditions.

Main Methods:

• Developed EffRes-DrowsyNet by integrating EfficientNetB0 and ResNet50 architectures.
• Utilized video-based analytics for fatigue detection.
• Conducted experiments on SUST-DDD, YawDD, and NTHU-DDD benchmark datasets.

Main Results:

• Achieved high accuracy (97.71%) on SUST-DDD, (92.73%) on YawDD, and (95.14%) on NTHU-DDD.
• Demonstrated superior generalization and classification performance in various settings.
• Obtained high precision (98.07%) and recall (97.33%) on SUST-DDD.

Conclusions:

• Hybrid deep learning models are effective for real-time, safety-critical applications like driver monitoring.
• EffRes-DrowsyNet offers a scalable and adaptable solution for attention-critical domains.
• The model shows promise for enhancing automotive safety systems and beyond.