Micro-sleep detection using combination of haar cascade and convolutional neural network

The level of traffic accidents is increasing every day. One of the factors causing traffic accidents is the condition of drivers who are tired and feeling drowsy. Referring to that problem, we propose an early warning system for detecting micro-sleep based on the image processing analysis using NUC processing. Micro-sleep is a condition when someone falls asleep but only a few seconds, usually between three seconds. A camera is used as a sensor that receiving an input image for obtaining information on the state of the eyes in the drowsy condition or not. An indication of a driver being sleepy or in the micro-sleep condition can be analyzed from the amount of blinking in his eyes. The combination of Haar Cascade and Convolution Neural Network is proposed. Haar Cascade will detect the face area and mark the eye region for the beginning of the program. Moreover, Convolutional Neural Network (CNN) is used to detect open and closed eyes. Detection of micro-sleep using a combination between Haar Cascade and Convolution Neural Network has an average accuracy of 97.23% when the condition of the user is 30-50 cm in front of the camera. In this system, the average computation time is obtained 0.2075 s.