| For fatigue driving,there will be certain safety hazards during the driving process of the vehicle.If not immediately reminded,the fatigue may be more serious and the life safety of the driver will be affected.Therefore,the driver’s fatigue detection and fatigue prediction are carried out during the driving process.It can effectively guarantee driving safety.In view of the problem of fatigue driving detection,researchers mainly propose three different ways to solve this problem: detection based on driver’s physiological characteristics,detection based on vehicle motion characteristics,and detection based on driver’s facial characteristics.Although among these three solutions,the results based on the driver’s facial features can best reflect the current driver’s fatigue state,and this solution is non-invasive,and only one camera is required to obtain facial features.Therefore,this solution is widely used.Applied in scientific research.This paper aims at the fact that face pictures are susceptible to the effects of light and head posture during vehicle driving.The Gabor feature and LBP feature of the image are combined,and the traditional image processing technology and deep learning are combined through transfer learning.In the fatigue recognition detection task,the data is video data,and the single extraction of spatial domain features from the image will lose the time domain features.Therefore,this paper proposes a fatigue recognition network combined with Bi LSTM.The main contents of this paper are as follows:(1)During the driving of the vehicle,the driver’s head will be bowed and tilted.Among them,the left and right deflection is the most frequent head position.When the driver tilts the head to the left and right,part of the face and eyes will appear.Occlusion has an impact on the feature extraction of the eyes.In daily life,the behavior of the eyes is generally synchronized.Therefore,an eye screening mechanism is proposed,which uses a single eye to replace the traditional eyes.(2)It is very important to reduce errors caused by illumination and improve the accuracy of fatigue driving detection.Therefore,it is designed to combine the Gabor features and LBP features of the facial,eye,and mouth region images,and finally merge the three features.Join the training set to train the network.(3)Combining traditional image processing technology with deep learning through transfer learning,a feature extraction network is proposed,and the global module(GModule),discrimination module(P-Module)and cross-channel are added on the basis of Dense Net.Pooling layer(pooling),and finally through the three results for weighted summation.(4)A CNN+Bi LSTM fatigue recognition method is proposed,which uses Bi LSTM to connect the final convolutional layer and the fully connected layer of the feature extraction network respectively to extract timing features,merge the two extraction results,and finally perform softmax Fatigue classification judgment. |
