Fatigue Detection of Driver using Convolutional Neural Networks

Abstract

ABSTRACT
Vehicular and road accidents claim millions of lives every year. In majority of these accidents, the driver is at fault. The severity of these incidents is hugely dependent on the temperament of the driver. Tired or fatigued drivers are usually less alert, and hence, are more prone to lose control and cause accidents. These circumstances warrant the need for development of systems to monitor the fatigue level of drivers and prevent hazardous accidents. Hence, we propose a system to detect fatigue or drowsiness level of drivers in real-time and take regulative action in the form of alarms to prevent accidents and injuries and avoid property damage. A deep learning-based approach is employed to make the system accurate, robust, and applicable in real-time. The system begins by processing a live video signal captured by a mounted camera. This video is converted to images and a real-time face detection algorithm is used to establish a region of interest and identify faces from the images. Next, feature extraction is performed on the identified faces by using a Convolutional Neural Network (CNN) model to detect important features for fatigue detection like the eyes, eyelids, mouth, etc. This model is trained for fatigue detection and acts as classifier to predict the fatigue level of the driver. If the fatigue level is high and persistent, an alarm will be sounded in the form of a warning to alert the driver. The complete process is carried out in a compact, inexpensive but powerful embedded board like raspberry pi.

Keywords: deep learning, convolutional neural networks, image processing, video processing, feature extraction, fatigue detection

Cite this Article: Manjusha Namewar, Mihir Rane, Sumedh Mhetre, Rahul Kumar. Fatigue Detection of Driver using Convolutional Neural Networks. International Journal of VLSI Design and Technology. 2020; 2(1): 24–30p.