A deep learning approach to sleep history classification during the task of driving
Georgia Tuckwell, A deep learning approach to sleep history classification during the task of driving
Winner – CQUniversity VYT local competition (2022)
Driving when fatigued as a result of inadequate sleep poses a significant risk to the all road users. Fatigue leads to reduced reaction time and alertness, and is a major factor in 20% of all road accidents worldwide. We developed a sleep history detection method using deep learning, which analysed driver movements as they control the vehicle, to classify if the driver had an adequate, or inadequate amount of sleep. Deep learning is a type of machine learning and artificial intelligence which learns from experience, in a similar manner to that of a human brain. We used an accelerometer, a tool which measures human movement from 3 axes, and attached it to the thigh which would control the brake and accelerator of the driving simulator. Using less than 4 minutes of accelerometer data, we were able to accurately classify sleep history 88% of the time! This approach opens-up many possibilities to help reduce the impact of fatigued drivers on the road.