Deep learning neural network for driver drowsiness detection using eyes recognition

Abstract

Driver drowsiness and fatigue are two of the most common causes of automobile accidents. Yearly, the number of deaths in addition to fatalities rises dramatically owing to a variety of factors. Therefore, a system must be found that alerts the driver to reduce traffic accidents and thus preserve people’s lives and public property. Within this research, Driver Drowsiness Detection System (DDDS) created on eyes movement, was designed and implemented to protect the driver from accidents .The proposed DDDS utilizes a method for identifying driver drowsiness that is based on the driver’s actions utilizing their visual features using eyes movement (closed or open) by using a high-resolution camera. The features of eyes images were extracted to be used as input to the Deep Cascaded Convolutional Neural Network (DCCNN). The DCCNN was constructed and practically implemented in Raspberry Pi microcontroller (model B3) to distinguish the face district, as a result, the issue of inaccuracy resulting from artificial feature extraction is avoided. The frontal driver Landmarks face in a frame from the Dlib toolkit was used in the current study. As maintained by the eyes Landmarks, an advanced factor, named Eyes Aspect Ratio (EAR) is explored for evaluating the fatigue of the driver within the existing frame of the eyes picture. Then, the output of DCCNN was used to alarm the drive case. In the experiment, a 450×320 eyes resolution image was adopted with 60 frames/second (f/s). The experimental results revealed that the drowsiness detection accuracy was 99% . The results of the current research outperformed some previous studies in terms of drowsiness detection accuracy.