ABSTRACT
To comprehensively consider the feature parameters for facial fatigue, to be able to fully characterize fatigue state, to improve the accuracy of driver fatigue recognition, and to improve vehicular safety and accident prevention, this paper proposes an improved local binary pattern feature extraction method to identify fatigue driving state using CNN deep learning model. A basic image dataset for fatigue recognition was constructed based on driver images under different lighting conditions. The acquisition and generalization of sample images of the dataset were completed through self-built dataset as well as pre-processing and data enhancement work. An 8×8 block-weighted LBP algorithm was proposed to extract the initial features of the images, and the facial feature texture of the driver was extracted from the dataset images as the input of the convolutional neural network to construct a CNN model for driver fatigue status recognition. The results showed that the CNN model has good recognition accuracy and generalization ability, and the accuracy rate of the model is 93.52%, with high recognition accuracy and good stability. It provides a new method for the recognition of driver fatigue state.
