EFFNet-CA: An Efficient Driver Distraction Detection Based on Multiscale Features Extractions and Channel Attention...

Area of Science:

• Computer Vision
• Artificial Intelligence
• Road Safety

Background:

• Driver distraction is a primary cause of road accidents, leading to injuries and fatalities.
• Existing deep learning methods for driver activity detection often suffer from high real-time false prediction rates.
• There is a critical need for effective real-time driver behavior detection systems to prevent accidents.

Purpose of the Study:

• To develop an effective and efficient real-time driver behavior detection technique.
• To improve the accuracy of driver distraction detection systems.
• To enhance road safety by minimizing accidents caused by distracted driving.

Main Methods:

• A convolutional neural network (CNN) model integrated with a channel attention (CA) mechanism was developed.
• The proposed model was compared against various backbone models (VGG16, ResNet50, Xception, InceptionV3, EfficientNetB0) with and without CA integration.
• Performance was evaluated using standard metrics like accuracy, precision, recall, and F1-score on the AUCD2 and SFD3 datasets.

Main Results:

• The proposed CNN+CA model achieved superior performance compared to baseline models.
• High accuracy rates were recorded: 99.58% on the State Farm Distracted Driver Detection (SFD3) dataset and 98.97% on the AUC Distracted Driver (AUCD2) dataset.
• The model demonstrated effectiveness in real-time driver behavior detection.

Conclusions:

• The developed CNN-based technique with channel attention offers an effective solution for real-time driver behavior detection.
• This approach shows significant potential for enhancing road safety and reducing accident rates.
• The model's high accuracy validates its capability in identifying distracted driving behaviors.