Research On Fatigue Detection Of Train Drivers Based On Facial Feature Fusion

With the rapid development of China ’s railway,China has the world ’s largest and most complex railway network.In the face of increasing workload and complex working conditions,the probability of fatigue driving of train drivers also increases.It is the most objective method to estimate the driver ’s fatigue state by analyzing facial features.However,for the face images existing in the real driving environment,the complex lighting environment,driver ’s line of sight changes and call response will affect the extraction of fatigue features.Aiming at the typical facial features of train drivers ’ fatigue driving,this paper proposes a fatigue detection method based on neural network facial multi-feature fusion.The main research contents are as follows:This paper first proposes an image contrast enhancement method to improve the contrast and clarity of the train driver ’s face image.According to the distribution characteristics of the illumination component of the face image,the parameters of the Gamma function are adjusted to realize the adaptive correction of the face image with uneven illumination.Secondly,the lightweight network structure Mobile Net is introduced to reduce the complexity of the model,and the network layer structure of MTCNN is re-adjusted.The improved network is used to train the face dataset images.The experimental results show that the Miou value is 94.8 %and the loss value is 0.3 %.After extracting the driver ’s fatigue characteristics,the k-means ++ method is used to select the adaptive threshold of the human eye opening degree,so that each driver has his own eye opening degree and improves the robustness of the detection method.Aiming at the error of the detected eye opening and closing degree when the driver bows or raises his head,the homography transform is used to correct the facial feature points,so that the correction result is consistent with the actual situation.Finally,fuzzy inference is used as a tool to formulate fuzzy rules,and the three indexes of eye opening degree,mouth opening degree and head posture angle are fused to judge the driver ’s fatigue state.The accuracy of this method is 97.8 %.In this paper,the MTCNN network model is improved to achieve the best balance between accuracy and detection speed.The designed facial feature fusion fatigue detection method has high accuracy,realizes fatigue classification and meets real-time requirements,and has practical application value.