The Indian Institute of Technology, Ropar, has developed an algorithm for driver drowsiness detection using machine learning and computer vision.
A study titled ‘Machine Learning Models for Drowsiness Detection’ jointly published by Harshit M and J.M.P. Ganesh of the Department of Mechanical Engineering and Dr. Ashish Sahani of the Center for Bio-Medical Engineering says they used computer vision algorithms to extract facial features such as eye closure and yawning, followed by machine learning techniques to effectively detect driver’s alertness.
It is an industrial and academic challenge to develop drowsiness detection technologies. Multiple techniques have been developed in recent years.
“One method is where the driver’s operation and vehicle behaviour can be monitored by the steering wheel movement, accelerator or brake patterns, vehicle speed, lateral acceleration and lateral displacement. Another set of techniques focuses on monitoring physiological characteristics of the driver such as heart rate, pulse rate, and electroencephalography. The third set is based on computer vision systems which can recognise the facial changes occurring during drowsiness. The first method is limited by the type and model of the car. The second method though with more accurate results has widely been downplayed due to impracticality in deploying it on a large-scale and its intrusive nature. The third method is a very promising one, which we have followed and developed a model on the same,” said the researchers.
“We have successfully developed an algorithm using machine learning and computer vision. We used hand-engineered features detecting drowsiness based on human facial expressions. We were able to come up with an effective solution, with little inconvenience caused to the driver in the form of close to body sensors and instruments. Also, it will work efficiently irrespective of the model and age of the car. We developed an algorithm to detect a drowsy state in real-time.”
The study said though the proposed machine-learning based detection can detect drowsiness with reasonable accuracy, there is still scope for improvement in its performance.