Research On Analysis And Identification Of Driving Dangerous Scenarios Based On Information Fusion

With the continuous development of China’s economy,the number of motor vehicles in China has also increased significantly.The issue on road traffic safety has attracted attention from public as well as scholars.Dangerous scenarios arise when vehicles collide with other road traffic users during driving,and it is of great importance.For inexperienced drivers,it can be used as an early warning which can avoid unnecessary accidents;For technical applications,dangerous scenarios can be used as a theoretical basis for driving assistance systems,thereby reducing traffic more scientifically.In this paper,the differences of PPG and EEG signals of drivers in different dangerous scenarios are analyzed and a dangerous scenario identification model based on support vector machine or random forest is established.Firstly,the mechanism and influencing factors of the dangerous scenarios are analyzed,and the evaluation methods of the dangerous scenarios are established.Then the physiological indicators which can characterize the driver’s mental state of the driver are selected.Different dangerous scenarios represents different stress levels and workloads of the driver.The real vehicle experiment is adopted to collect the PPG and EEG signals of the driver in different dangerous situations.Five different dangerous scenarios are extracted and counted by real vehicle video,and the corresponding physiological data is obtained.Heart rate data is obtained by preprocessing PPG data using Ergo LAB;EEG data is exported using EEGLAB plug-in,and pre-processing EEG data.Then,heart rate growth rate and heart rate variability(time domain and frequency domain)analysis are performed on the driver’s PPG signal;the driver’s EEG signal is further removed by wavelet transform,and the power of ? wave,? wave,? wave and ?/ ? in the EEG signal the spectral density is extracted and the differences of various waveforms and indicators in different dangerous scenarios are analyzed.Finally,a multi-source information fusion method is used to construct a dangerous scenarios identification model based on support vector machine.Principal component analysis is performed on the fused indicator matrix,and seven eigenvalues with cumulative contribution rate ≥90% are retained.Construct a support vector machine model of one-to-many classifier and a random forest model.Libsvm is used to optimize the parameters by particle swarm optimization.After obtaining parameters,the model which is trained to make predictions is compared to the single data source(PPG and EEG)before fusion.The classification accuracy of the model can reach 82.01%.And compared with the classification accuracy of single data source(PPG or EEG feature)before fusion.The classification accuracy of the random forest identification model can reach 90.65%,which shows the great validity and applicability of the model.