Driver Distraction Detection based on Deep Learning Techniques using Images

The alarming increase in road fatalities demands immediate action. In 2019 alone, over 1.35 million lives were tragically lost on the world’s highways. Addressing distracted driving, a key contributor to this surge, is crucial for accident prevention. While prior research has explored Deep Learning for distracted driver detection in vehicles, there is still room for advancement in identifying drivers whose focus is diverted from the road, a specific type of distraction. This study proposes a Deep Learning-based classification system to detect drivers looking away from the road. It leverages Jupyter Notebook and Python to implement and evaluate ResNet50, VGG16, and InceptionV3 network architectures. The State Farm dataset, encompassing ten distinct categories of driver behavior, is utilized for training and testing. These categories include actions like talking on the phone, texting while driving, eating, drinking, reaching behind for objects, adjusting the radio, grooming, talking to passengers, and looking elsewhere. The focus of this study is on improving detection accuracy specifically for the "looking elsewhere" class. Model performance is assessed using confusion matrices and accuracy metrics. The results reveal that the ResNet50 model achieved a noteworthy 95% accuracy in classifying drivers looking elsewhere. VGG16 and Inception models also exhibited strong performance, reaching accuracies of 89% and 92% respectively.