Unusual Driver Behavior Detection in Videos Using Deep Learning Models.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Road accidents are a major concern, with abnormal driving behaviors contributing significantly.
• Real-time monitoring of driver actions is crucial for enhancing safety and preventing accidents.
• Automated detection systems offer a more effective and rapid solution compared to continuous manual monitoring.

Purpose of the Study:

• To develop and evaluate deep learning models for automated detection of anomalous driving behaviors.
• To identify common distracting activities such as smoking, eating, drinking, and calling while driving.
• To compare the performance of a proposed Convolutional Neural Network (CNN) model against established pre-trained models.

Main Methods:

• A novel dataset was created comprising five classes: Driver-smoking, Driver-eating, Driver-drinking, Driver-calling, and Driver-normal.
• Deep learning models, including pre-trained CNNs (ResNet101, VGG-16, VGG-19, Inception-v3) and a proposed CNN-based model, were trained and tested.
• Performance evaluation was conducted using standard metrics to compare the effectiveness of different models.

Main Results:

• The proposed CNN-based model achieved a classification accuracy of 95%.
• Pre-trained models demonstrated strong performance, with accuracies ranging from 89% to 94%.
• The proposed model outperformed the pre-trained models in classifying abnormal driver actions.

Conclusions:

• The developed CNN-based model is highly effective for detecting abnormal driver behaviors.
• This automated system can significantly contribute to improving driver safety and reducing road accidents.
• The findings highlight the potential of deep learning in real-time driver monitoring applications.