Research On Fatigue Detection Of Controllers’ Facial Features

As an important guarantee personnel for the safe operation of aircraft,the air traffic controller’s fatigue detection research is one of the important topics in the field of aviation safety.Real-time detection during on-the-job work is critical.Considering the actual working conditions and environment of controllers,this paper proposes a controller fatigue recognition algorithm based on facial features.The algorithm includes face detection module,feature extraction module and fatigue evaluation module.First,the face detection module detects faces and eyes through a multi-task cascaded convolutional neural network.Secondly,the feature extraction module mainly extracts features through the fatigue expression recognition model and the eye state recognition model to identify the fatigue state of a single frame picture.The controller fatigue expression recognition model is an improvement of the Xception neural network,and adopts the transfer learning strategy to fine-tune the fatigue expression recognition model built on the fatigue expression dataset to obtain the model and weights.The controller’s eye state recognition model adopts the transfer learning strategy to pre-train the deep neural network and the deep convolutional neural network,and fuse the networks to obtain the eye state recognition model.Finally,the fatigue evaluation module is the result of the fusion of two static fatigue identification models.According to the controller fatigue judgment index,the controller fatigue characteristics are established.And based on the improved D-S evidence theory,the decision-level fusion of FFE,MDFE,PERCLOS,MDEC and BF is carried out to comprehensively determine the controller status.The average accuracy rate reached 94.5% after testing the front-line controllers who performed recurrent training tasks.It has a high accuracy rate for controller fatigue detection,which helps to reduce the occurrence of controller fatigue on the job.