Fatigue driving detection method based on multi-task cascaded convolutional network

Abstract

Fatigue driving has become an important cause of traffic accidents, but fatigue driving detection methods are still in the preliminary stage of development. Detection methods based on driver’s physiological characteristics are expensive, and detection methods based on vehicle motion characteristics are low in accuracy. In order to overcome the difficult problem of fatigue driving detection, the paper proposes a method based on multi-task cascaded neural network (MTCNN), which uses key point coordinate detection of human face, three courtyards and five eyes to locate human eyes, eye aspect ratio (EAR) to detect fatigue and other methods to detect fatigue driving with high accuracy, non- invasive, low cost and other advantages. The experimental results show that the proposed scheme has improved detection accuracy and reduced detection time compared with Convolutional Neural Networks (CNN), which provides reliable technical support for the research on fatigue driving detection.