A Study On Driver Cognitive Distraction Detection Based On The Driving Performance Measures

In recent days,with the rapid development of information technology,driver distraction has become an important factor for road traffic accidents.How to monitor driver's status while driving in real time has become the focus of the traffic safety field.Driver distractions are mainly divided into two types: visual distraction and cognitive distraction.This paper aims at the driver cognitive distraction,mainly completed the works as follows:(1)Design experimental scenario by Carsim software,and the secondary tasks according to the reported cognitive distracted event in the SHRP2 to increase the cognitive workload of drivers and impose them to cognitive distraction.The eye movement data and driving performance data were synchronously collected in driver-in-Loop simulator,and then were preprocessed using Pauta(3?)criterion method for eliminating outliers and Z-score method for standardization,finally,the database for detection of driver cognitive distraction were established.(2)Extract measures which were great useful for detecting driver cognitive by using the Relief algorithm and principal component analysis.For purpose of modeling,it's essential to define class labels.This study proposed using the selected eye movement data for labeling the corresponding driving performance measures.And these labeled driving performance data were used for training and testing SVM models to recognize driver cognitive distraction and explore the influence of different experimental road and secondary tasks.The results showed that the driver's driving behavior changed more significantly when he we re asked to perform the continuous subtraction calculation tasks while driving in curve segment of experimental road.Moreover,the BP neural network models were constructed in the same training and testing data sets with SVM models,and the results of the comparison between the classification performances showed that the BP neural network were outperform the SVM models in detecting cognitive distraction with an average accuracy of 87.69% under different experimental conditions.(3)Construct BP neural network model with data collected from the experiment with phone-related secondary task to detect driver cognitive distraction in real time,and then explore the influence of the different combinations of time window size and overlap for detection of cognitive distraction.The results showed that the best average accuracy of the BP neural network model trained from the input data with the combination of 60-s window size and 95% overlap was 85.2%.