Research On Driver’s Physiological Features And The Effect Of Warning Based On Vehicle-pedestrian Conflict Event In Visual Blind Area

Vehicle-pedestrian collision accidents occur frequently on roads,which is an important reason for the high number of pedestrian casualties.Accidents caused by blind spots are more difficult to avoid due to the emergency.In vehicle-pedestrian conflict accidents,the final collision avoidance result is closely related to drivers’ perception-decision-reaction process.Therefore,exploring drivers’ cognitive process in vehicle-pedestrian conflicting accidents from perspectives of physiology and behavioral performance is crucial to reducing traffic accidents and ensuring road traffic safety.Based on the Forward-Collision-Warning technology in the Io V environment,a driving simulation experiment of vehicle-pedestrian collision accidents in the visual blind spot was designed using the driving simulator.Roadsides parking was selected as the basic scene of this experiment,and different road speed limits,pedestrian speeds and warning release times were considered to explore the effect of different factors on driving behavior.During the experiment,driver’s behavior,eye movement,and EEG data were collected,and the whole collision avoidance process was divided into approach stage,perception stage and reaction stage.Then some key variables were extracted from behavior data,eye movement trajectory and EEG signals,which were analyzed by some statistical methods such as descriptive statistics,mixed-effects model,correlation analysis and logistic regression.Finally,three machine learning algorithms were adopted respectively to establish prediction models of vehicle-pedestrian collisions,while external variables,behavioral variables,eye movement features and EEG features significantly related to the collision avoidance results were selected as input variables.A variety of evaluation indicators were selected to evaluate performance of different models.This study found that the speed limit,pedestrian speed and release time of warning had significant effects on collision-avoidance performance.At the physiological level,drivers’ eye movement variables(gaze,blink,saccade,pupil diameter,etc.)and the power spectral density of different EEG bands were closely related to drivers’ behavioral performance.Physiological features in the perception stage could reflect drivers’ cognitive ability,which was crucial for successful collision avoidance.Besides,the warning releasing times could actively intervene in drivers’ cognitive process in the perception stage,which helped to improve drivers’ risk perception ability and significantly reduced the collision rate.From the comparison of different evaluation indicators of SVM,Random Forest and Ada Boost models,the model constructed by the Ada Boost algorithm performed better than the other two models.Eye movement variables and EEG features in the perception stage were more reliable as input variables of the prediction model.In general,the analysis conclusion of the multi-dimensional experimental data and the applies of various algorithms in the prediction of vehicle-pedestrian collision can provide a theoretical basis for the improvement of Forward-Collision-Warning technology,and expand new ideas for applications of eye-tracking and EEG technology in other traffic scenarios in the future.There are 59 figures,32 tables and 200 references.