Whether we are driving or in a passenger seat, we have seen drivers being distracted by a text on a phone, reaching behind for something, or having to attend a phone call.
According to U.S Department of Transportation, Every day about 8 people in the United States are killed in crashes that are reported to involve a distracted driver.
We think that having a system that alerts the driver if they are distracted, is the need of the hour.
We intend to do this using Convolutional Neural Networks and classify the drivers actions into 10 classes- safe driving, texting - right,talking on the phone - right, texting - left, talking on the phone - left, operating the radio, drinking, reaching behind, hair and makeup, talking to passenger.
DATASET - State Farm Distracted Driver Dataset
The contents of the notebook can be found below.
Part 1: Data exploration
Part 2: CNN model built and trained by us.
Part 3: EfficientNetB3 fine-tuned (Transfer learning)
Part 4: Visualisations
Part 5: Error Analysis
Part 6: Testing the model against custom test data