Distracted Driving State Discrimination And Safety Risk Prediction Based On Multiple Data Sources

The complex road traffic system consists of four elements: people,vehicles,roads and the environment,and most of the factors that contribute to all road traffic accidents are related to drivers,according to the data,the percentage of traffic accidents caused by distracted driving in China reached 38%.With the popularity and application of intelligent terminals and in-vehicle electronics,the hazards of distracted driving to traffic safety have become a pressing problem and how to detect distracted driving has become the focus of attention in the field of traffic safety,so it is necessary to carry out research on distracted driving.In this paper,distraction experiments are designed for young drivers who drive distractedly,and experiments are carried out through driving simulators.Driving performance data and EEG signal data of different driving states are extracted,and a driving distraction discrimination model and a collision risk prediction model are constructed.It provides theoretical support for the development and application of future distraction monitoring systems and distraction warning systems,and is of great practical significance to reduce the safety hazards caused by distracted drivers.The thesis mainly accomplishes the following work.(1)Design of driving distraction experiments.A distracted driving road scenario was built based on UC-win/Road software,using the "arrow task" and the "n-back task" to replace visual and cognitive distractions respectively,driver’s driving performance data and EEG data are collected during the experiment.(2)Analysis of the effects of distracted driving on drivers’ manoeuvring behaviour and EEG signals.After pre-processing the data,a one-way ANOVA was used for significance analysis.For the driving performance data,the differences between the different distraction states were compared in terms of lateral control behaviour and longitudinal control behaviour respectively;for the EEG signals,the power spectra of different frequency bands of waves in different brain regions were extracted using power spectroscopy to analyse the changes in brain activity state under different distraction states.(3)Construction of a multi-data source driving distraction discrimination model.The SVM and CNN-LSTM driving distraction discrimination models were built based on driving performance data and EEG data respectively,and the performance of CNC-LSTM was evaluated based on the performance of the models to derive the discrimination effect of CNC-LSTM.Driving performance data and EEG data were separately used as inputs to the CNN-LSTM model to judge the discrimination effect of different data sources,and the results concluded that driving performance data had better effect.(4)Correlation study between driving behaviour and brain activity state.Based on the driving performance data during the deceleration and collision avoidance phase and the EEG signals before deceleration,the correlation between driving behaviour and brain activity state was explored in terms of brain areas,EEG frequency bands and driving behaviour data,respectively.(5)Construction of a crash risk prediction model.The random forest model was used to rank the importance of the features of the driving behaviour data in the deceleration and collision avoidance phase,and the top six indicators and EEG data were used as inputs to the model to construct the collision risk prediction model,it was found that the model with driving performance data and EEG data predicted better in comparison to the crash prediction model with only driving performance.