Safety Behavior Evaluation Of Existing Railway Train Dispatchers Based On Eye Characteristics

The existing railway traffic controllers have heavy workloads and long working hours,which will affect their safety behavior to a certain extent.Evaluating and identifying the safety behavior of railway traffic controllers on existing lines during their on-the-job period and giving them a warning in time in an unsafe state is a significant problem to ensure the safety and efficiency of railway traffic.In the study of relevant literature,this paper found that the safety behavior can be reflected in the characteristics of fatigue and emotion,and that there is a correlation between fatigue and lousy emotion and the safety behavior of dispatchers in the data during the experiment.The fatigue state does not have the recoverability of bad emotional state,and the degree of fatigue will gradually increase.Therefore,aiming at the problem of safety behavior evaluation of existing railway traffic dispatchers,this paper proposes an unsafe behavior identification model based on the characteristics of dispatchers’ eyes and a time series prediction model of dispatchers’ fatigue degree.The former records five indexes of the eye characteristic information of the existing railway traffic dispatcher at work: fixation time,average pupil size,eye closure time,blink frequency,and saccade speed,and takes them as the input of the model.Before the dispatcher works,the dispatcher enters irritability,boredom,tension,and sleep deprivation respectively through the experiments of "video emotion induction" and "sleep deprivation," while four fatigue states are used as the output of the model.The latter takes the five eye movement indexes of the dispatcher in the fatigue group as the input of the model and extracts the fatigue degree value(i.e.,the average height-width ratio of both eyes)of the dispatcher in the working process through face feature recognition as the output of the model.To improve the accuracy of model evaluation under different safety behaviors and the accuracy of dispatcher fatigue value prediction model,this paper carries out real-time facial expression recognition for dispatchers in the process of emotion induction,ensuring the effectiveness of input data as much as possible.Moreover,K-means clustering is used for the average height-width ratio of both eyes to eliminate eye movement index segments with obvious clustering differences in the corresponding period of the fatigue group.In terms of the model,this paper establishes the safety behavior recognition model of existing railway traffic dispatchers based on the KNN algorithm,CHAID decision tree,and cart decision tree algorithm,respectively.And then,this paper establishes the time series prediction model of fatigue degree of existing railway traffic dispatcher based on Hidden Markov algorithm,CNN,LSTM,GRU,and CNN-GRU fusion algorithm.Then,the final model is determined according to multiple evaluation indexes.The results show that the average pupil size,gaze time,saccade speed,blink frequency,and saccade rate of dispatchers under the four unsafe working conditions are significantly different.Besides,the recognition accuracy of KNN model can reach more than 90%.The dispatcher fatigue prediction model based on the CNN-GRU fusion algorithm has the best fitting effect,and the goodness of fit R2 reaches0.8446.Therefore,the KNN model and CNN-GRU fusion model are used as the safety behavior evaluation model of the existing railway traffic dispatcher.