Research On Locomotive Drivers’ Fatigue State Detection Based On Upper Body Postures

Drivers must keep a fixed sitting posture and maintain concentration during the train operation.This process will make people fatigued easily.When a locomotive driver is fatigued,it is difficult for him to concentrate,which will delay the response and result in hidden dangers to driving safety.On the premise of normal train operation without interference,fast and accurate detection of the locomotive driver’s state without contact is crucial to driving safety.Although there are many current studies on contactless locomotive drivers’ fatigue state detection,these studies have used little body posture information,and have not unearthed the deep correlation between body posture information and drivers’ fatigue state.Taking the detection of the driver’s fatigue state based on upper body postures as the research object,the laboratory virtual simulation environment was used in this thesis to study the deep correlation between body postures and fatigue state in driving scenarios.Two detection methods of the fatigue state based on deep learning,as well as a two-stream fusion network model were proposed.The specific research content includes the following aspects:(1)As for the detection of drivers’ fatigue state,this thesis analyzed the research results at home and abroad.It introduced the high-resolution network of the body posture detection framework and explained the basic theories of Deep Belief Network(DBN)and Long Short-Term Memory(LSTM).It also analyzed the evaluation indicators of body fatigue and clarified the research content,laying a theoretical foundation for this thesis.(2)Considering the characteristics of the data structure,a DBN model good at extracting high-level distribution features and an LSTM model able to make use of data spatiotemporal information were constructed in this thesis to detect the fatigue state of locomotive drivers based on upper body postures.The effects of these two models were analyzed based on the sorted data.(3)A method of optimizing the network structures of the DBN model and the LSTM model based on the improved particle swarm algorithm based on genetic algorithms was proposed.The optimized model was compared with some traditional models so as to highlight their advantages in terms of model accuracy,recall,precision and F1 score.Then,a two-stream fusion network integrating the DBN model and the LSTM model was further studied.The structure design,parameter optimization and training of the model,as well as comparative testing and analysis were carried out.(4)As for the practical application of the fatigue state detection software for locomotive drivers,six functional modules in accordance with software requirements were designed.Based on the Python programming language,a real-time fatigue state detection system for locomotive drivers based on upper body postures was developed on the basis of the cloud platform.The functions of the software were verified.The research results showed that in the laboratory environment,the accuracy of the two-stream fusion network model proposed in this thesis reached 95.67%,and the macro F1 score reached 95.70%.Compared with other models and driver fatigue state detection methods,its overall effect has significant advantages with good robustness.It can better identify the deep correlation between upper body postures and drivers’ fatigue state,which is suitable for the fatigue state detection of locomotive drivers based on upper body postures.Figures 66,Tables 13,References 68.