Driver Drowsiness Detection Using Convolutional Neural Networks

The presented paper deals with automatic detection of driver drowsiness. Detecting the driver's drowsiness behind the steering wheel and then alerting him may reduce road accidents. Drowsiness in this case is captured using an autocamera, whereby, based on the captured image, the neural network recognizes whether the driver is awake or tired. The convolutional neural network (CNN) technology has been used as a component of a neural network, where each frame is evaluated separately and the average of the last 20 frames is evaluated, which corresponds to approximately one second in the training and test dataset. First, we analyze methods of image segmentation, and develop a model based on convolutional neural networks. Using an annotated dataset of more than 2000 image slices we train and test the segmentation network to extract the driver emotional status from the images.