Distracted driving is an epidemic in the United States; every day about 1,000 people are injured and nine are killed from car accidents involving distracted driving [1]. Thanks to a data set collected from a State Farm Machine Learning Competition, we have access to dash-cam footage of safe and unsafe driving. Our project aims to use this dataset to help identify distracted driving with computer vision.
For this project, we specifically worked with a basic 4 layer CNN and a pretrained VGG-16 model architecture. Below are conceptual images of their design along with the relative losses found during training.
Summarized results from our implementation.
