Research On Driving Fatigue Detection Model Based On Multi-feature Fusion Of EEG And Eye State Analysis

Traffic safety is a crucial component of social development,directly related to the safety of people’s lives and property.As cars have become an essential mode of transportation in daily life,road safety issues have gained increasing attention.Among these issues,driver fatigue is a significant factor contributing to traffic accidents.Driver fatigue notably reduces the driver’s reaction time and judgment,thereby increasing the risk of accidents.Therefore,researching and applying driver fatigue detection technology is of great importance for traffic safety management.This thesis aims to establish an efficient and stable driver fatigue detection model to achieve consistent detection of driver fatigue characteristics.The study investigates driver fatigue detection methods from the past decade,including active and passive detection methods.Existing detection methods face three key challenges.First,electroencephalogram(EEG)signals possess characteristics such as being non-static,non-stationary,non-linear,and having a low signal-to-noise ratio.The challenge is how to use EEG signals collected from drivers to construct effective and stable driver fatigue detection features.Second,EEG signals are time-series data,meaning that the signal at each moment may be influenced by the preceding moment.Extracting local features from EEG signals based on these characteristics makes it difficult to overcome the low signal-to-noise ratio and ensure the stability of application performance.Third,physiological characteristics and behaviors of different drivers in a fatigued state may vary.Relying on single-modal features may fail to meet the needs of real-world,complex,and variable application scenarios,potentially causing driver fatigue detection models based on a single modality to be ineffective.This thesis conducts in-depth research and exploration on the three key issues mentioned above,with the main research content and findings as follows:Firstly,research was conducted on the extraction of local features from EEG signals,proposing a multiscale entropy extraction method based on empirical mode decomposition under local electrodes,and presenting a fatigue driving detection model based on this entropy feature.Initially,comprehensive verification studies were conducted on four types of entropy applied to fatigue driving detection in a dataset of 26 subjects,further confirming that using only single-scale entropy features or multiscale entropy features that do not consider the influence of EEG frequency scales is difficult to effectively improve the performance of fatigue driving detection.Secondly,a study on the distribution of fatigue brain areas based on four types of entropy was conducted on the dataset of 26 subjects,confirming that different brain regions exhibit different EEG signal characteristics in fatigue state,providing evidence for the selection of local electrodes needed for fatigue driving detection.Additionally,to reduce time complexity and adapt to practical application scenarios,four electrodes were selected for EMD multiscale entropy feature extraction and fatigue driving detection model research.Thirdly,in order to more effectively obtain entropy features of EEG signals for fatigue driving detection,research was conducted on multiscale entropy feature extraction methods under local electrodes using EMD,specifically proposing the extraction of EMD multiscale entropy for EEG signals under local electrodes based on appropriate scale factors and the number of intrinsic mode function(IMF)components found during EEG signal preprocessing.The aim was to enhance the accuracy of fatigue driving detection and effectively balance the performance of fatigue driving detection and time efficiency.Fourthly,an analysis of EMD multiscale entropy features under local electrodes and experiments on fatigue driving detection performance were conducted on the dataset of 26 subjects to verify the effectiveness of this entropy feature in fatigue driving detection.Through comprehensive analysis of entropy features and fatigue driving detection experiments in fatigue driving scenarios,the validity of EMD multiscale entropy under local electrodes for fatigue driving detection was verified.Finally,experiments on the fatigue driving detection performance of EMD multiscale entropy features under local electrodes were conducted for individual data of 26 subjects,demonstrating the limitations of solely selecting local entropy features for fatigue driving detection.Secondly,research was conducted on the extraction of global features from EEG signals,and a method for constructing a functional brain network(FBN)based on frequency band shortest path tree/combined shortest path tree(SPT)was proposed.A driver fatigue detection model based on this functional brain network was also developed.Addressing the limitations of the EMD multi-scale entropy features under local electrodes in terms of individual variability and model stability,the thesis explored the construction of FBN using EEG signals and the extraction of global features,aiming to enhance the stability and accuracy of driver fatigue detection.Initially,building on the previously studied shortest path tree FBN construction,and considering the significant correlation between fatigue state and EEG signal frequency bands,a method for constructing a frequency band-based shortest path tree/combined shortest path tree FBN was proposed,including the determination of the root node of the shortest path tree.Secondly,driver fatigue detection models based on the frequency band shortest path tree and combined shortest path tree were constructed and subjected to thorough validation experiments.These experiments included verifying the effectiveness of driver fatigue detection based on shortest path trees and combined shortest path trees under different frequency bands and frequency band combinations,as well as comparing the effectiveness of different FBNs in driver fatigue detection.Finally,noise resistance verification was conducted for the driver fatigue detection based on the frequency band combined shortest path tree.Preliminary results indicated that the model possesses a certain degree of noise resistance and detection stability.Through extensive experiments,the effectiveness of the proposed EMD multi-scale entropy under local electrodes for driver fatigue detection was validated.Thirdly,research was conducted on the multimodal fusion of EEG signal features and eye movement features,and a driver fatigue detection method based on multimodal fusion features was proposed.Considering the complexity of real-world application scenarios and the fact that fatigue is a progressive process characterized by complex multi-feature manifestations,the thesis explored the fusion of multimodal features to enhance the adaptability and stability of the driver fatigue detection model.It proposed a driver fatigue detection method based on the fusion of EEG local and global features with eye movement features.Initially,eye movement feature extraction was conducted,and validation research on its effectiveness in driver fatigue detection was performed,confirming that eye movement features can be utilized for driver fatigue detection,while analyzing their advantages and limitations.Secondly,experiments were designed for synchronous feature extraction,data analysis,and fusion strategies for multimodal feature fusion.Finally,after conducting effective validation and result analysis of the driver fatigue detection model based on the fusion of EEG local EMD multi-scale entropy features and global features based on frequency band combined shortest path trees,the effectiveness of driver fatigue detection based on the fusion of two modalities(EEG and eye movement)with three features was fully validated and analyzed on a dataset of six subjects.Through extensive experiments,the effectiveness of the proposed driver fatigue detection model based on the fusion of EEG and eye movement features was validated.This model overcomes the limitations of single-modal features,improves the accuracy and stability of driver fatigue detection,and provides technical support for conducting driver fatigue detection in complex environments.The research findings of the thesis not only contribute to the scientific understanding of the brain-behavior relationship but also provide important evidence for the subsequent personalized driver fatigue detection and the improvement of system accuracy and universality.They also offer crucial guidance for policy-making in traffic management departments and present a new technical solution for enhancing driving safety and reducing the occurrence of traffic accidents.