Driver Distraction Detection System with Categorized Alerting based on SVM and PCA Optimization

The road accident scenario in India can be understood by simple statistics, and that every year more than 1.5 Lakh people lose their lives on the road. This is like crashing a jumbo jet carrying 340 people every day without any survivors. This is the enormity of the situation. Road accidents impact India’s GDP by 3.14% (IIT Delhi-TRIPP Study). This paper mainly focuses on the solution to the issue of distracted driving. Distracted driving is an act of indulging in attention-breaking activities while driving that can disrupt the flow of the driver’s coordination on the road. These distractions can be broadly classified as Manual, Visual, and Cognitive. The following research work shows how driver distractions can be detected using computer vision equipped cameras in vehicles to detect an act of distraction. We have used the State Farm Distracted Driver Detection Dataset (SFDDD), which consists of a set of images classified into 9 categories of driver distraction activities. We have adopted a Machine Learning approach and have trained a classification model using Linear Support Vector Machine (SVM), achieving an accuracy of 98.95%. The result has a trade-off with the time and space and was further improved by merging SVM with Principal Component Analysis (PCA) that significantly reduced the model training time and even enhanced the accuracy to 99%. Using the model, we have also implemented a prototype for prioritized alerting system that warns the driver as soon as it detects a distraction and sets off an alarm.