Research On Controller Fatigue Detection Algorithm Based On Deep Learning And Facial Multi Information Feature Fusion

Air traffic controller fatigue risk management is an important research of civil aviation safety in China.Research shows that controllers fatigue can directly cause series of accident syndromes,or even accidents.Therefore,the research on controller fatigue detection algorithm is greatly significance to civil aviation safety.With the development of artificial intelligence technology,deep learning which is used to detect controller fatigue has become possible.This paper is based on deep learning and facial multi information.The controller fatigue state is detected and analyzed and also the fatigue warning judgment is made.The main research is followed:(1)The algorithm of face image processing and face detection is studied.Aiming at the problems of dim light in approach control hall and high light intensity in tower control room,a method based on image decomposition and MSR illumination compensation is proposed in this paper.This method can preserve the image details and balance the image,which provides a good image input basis for subsequent facial feature detection.A fast face detection algorithm based on three-stage cascade convolutional neural network is studied.The algorithm uses regression frame to reduce the number of network structure,uses full convolutional neural network to identify face candidate frame,and uses global average pooling method to replace the traditional full connection layer,so as to improve the detection speed of network and reduce the loss of network parameters.In addition,in order to locate the face accurately,this paper uses the location confidence method to improve the traditional non-maximum suppression algorithm.(2)The algorithm of face key point detection and head pose estimation is studied.Based on the idea of multi task deep learning,this paper designs a feature extraction method of micro unit structure,which shares a network of facial feature points and head posture position for joint learning,and improves the detection accuracy of key points under the change of body posture;at the same time,it can also get accurate head posture estimation without increasing computational redundancy.(3)A controller fatigue evaluation system based on multi information feature index fusion is proposed.According to the description information of controller fatigue characteristics,the distributed information fusion structure is adopted.Based on improved rough theory,the controller fatigue information fusion judgment is carried,and the fatigue decision scale is constructed to judge the controller fatigue degree.(4)The controller fatigue detection system is developed.In this paper proposed the fatigue detection algorithm,also the fatigue detection system is developed on the framework of Jetson Xavier.The program is developed with Python on the idea of modular design,also the opencv is used on the image interface,and the fatigue detection is realized based on the core algorithm module.The system can monitor the fatigue state of controllers and give fatigue tips in current time.The algorithm of controller fatigue detection based on deep learning and facial multi information feature fusion can provide new ideas and theoretical guidance for the subsequent controller fatigue research.