Blink Fatigue Detection Algorithm Based on Improved Lenet-5

Abstract

Fatigue driving is the main factor in many traffic accidents. The eye behavior is the main direction in the field of fatigue driving research. It reflects the degree of fatigue of the human brain to some extent. In this paper, the publicized CEW human eye opening and closing data set is used to preprocess the human eye image, and then the preprocessed image is placed in the improved LeNet-5 convolutional neural network. In order to fully extract the human eye features, the network is added. The number of layers, at the same time, to speed up the network convergence speed in order to prevent the gradient from disappearing, the activation function is changed from Tanh to ReLU function. Experiments show that the algorithm has a blink recognition rate of 93.5% in the public CEW dataset, and an accuracy rate of 5.1% compared with the unmodified LeNet-5. This method has a good blink detection effect and has important application value in the field of fatigue driving. Keywords-Fatigue Driving; Blinking Algorithm; Convolutional Neural Network; Network Optimization