Research On Driver’s Eye Movement For Road-traffic-conflict Discrimination

With the rapid increase of vehicle population in China, road traffic safety has become a problem of concern. Traffic conflict technology, as a kind of traffic safety evaluation technology with no need of traffic accident quality, has shown a lot of advantages. Thus, the traffic conflict technology has become the mainstream of road traffic safety evaluation method. But the academic circles have not reached the agreement in the traffic conflict discrimination indicators and the traffic conflict discrimination method, the road traffic safety evaluation method has been hindered to a certain extent. The research of eye movement characteristics in traffic field shows that there is a close relationship between driver’s cognitive and eye movement characteristics, which provides a new direction to explore the method of traffic conflict discrimination with high consistency of drivers.This paper explores the existing problems and possible solutions of traffic conflict discrimination technology by analyzing current situation of research on traffic conflict technology and eye movement fields in China and abroad to explore the discrimination indicators and methods of traffic conflict that can be widely accepted.Aiming at existing problems, the eye movement indicators are analyzed, and the road traffic conflict test is designed to provide basic data for the research of conflict identification indicators and methods. The characteristics of eye movement indicators such as pupil diameter, eye movement and eye gaze position were studied using inferential statistical methods and descriptive statistical methods. On the basis of driver’s eye movement characteristics, the optimization algorithm and classification model are selected. Road traffic conflict discrimination method using gravitational search algorithm and support vector machine with driver’s eye movement indicators as input is proposed, and the discrimination method’s effect is verified. The specific research work of the road-traffic-conflict discrimination method based on driver’s eye movement is as follows.1) Current situations of the domestic and foreign research on the traffic conflict technology and eye movement in the traffic field are analyzed. Through the analysis of the traffic conflict technology research situation, the reason for the low consistency of drivers and traffic conflict technology research results is that the selected traffic conflict identification indicators are not based on driver’s cognitive. Therefore, it’sneeded to explore conflict discrimination indicators and method from driver’s cognitive. It is found that, by analyzing current eye movement research situation in traffic field, a certain research foundation is accumulated in the relationship between driver’s eye movement and cognition and selection of eye movement indicators, but the eye movement research did not involve the traffic conflict technology. It is considered that, it is feasible to start with the road-traffic-conflict which has relatively sample relationship, to explore the road-traffic-conflict discrimination method on the basis of eye movement characteristics and the relationship between eye movement and cognition and mental load.2) The eye movement indicators are analyzed and the road-traffic-conflict tests are designed. In the face of the research needs, which can be operated and quantified,the test collection indicators was selected through analyzing eye movement, fully considering road traffic environment changes and driver’s condition changes, to provide basic data for the research of conflict discrimination indicators and conflict discrimination method. The road-traffic-conflict test design is completed in full consideration of the characteristics of the variation of the test variables in the road-traffic-conflict, according to the control principle, the principle of the single factor variable, the principle of parallel repetition. According to characteristics, test data is preprocessed using the EMD and other methods.3) Driver’s eye movement in the road-traffic-conflict is researched. It is considered that, road-traffic-conflict will cause cognitive load and psychological pressure on the driver. In order to verify this hypothesis, firstly, the mean value of pupil diameter, the standard deviation of pupil diameter, the blink interval and the blink time are defined according to the data characteristics of pupil diameter and blink action indicators, and driver’s fixation points are defined according to fixation points’ characteristics by dividing driver’s areas of interest using the DBSCAN clustering method which can avoid rigid boundary, falling into the local optimum, or other defects of conventional clustering methods. Influence of illumination, trample movement, conflict time, conflict severity, type of conflict and other variables on the three types of eye movement indicators of pupil diameter, blink action and fixation points in the environment of road-traffic-conflict is defined using inferential statistics methods. Eye movement indicators which are significantly correlative with road-traffic-conflict are descriptive statistics analyzed, in order to get change laws of the relevant indicators with conflict severity and conflict type. The change characteristics of related eye movement indicators were defined through analyzingdescriptive statistical results, and the foundation for the conflict time positioning,conflict severity discrimination and conflict type discrimination based on eye movement is laid.4) The discrimination method for the road-traffic-conflict based on eye movement is studied. In order to solve the problem of large amount of data in the conflict period, the eye movement indicators representated key points of the conflict are defined according to change laws of eye movement in road-traffic-conflict, and a movement characteristics based and rapid location method for road-traffic-conflict is proposed drawing on the GSA optimization strategy. The current situation that the eye movement indicators have no obvious linear law and there’s no priori eye movement knowledge is fully considered. The SVM’s advantages that supervisedly learning data without prior knowledge and generating non linear model are analyzed. The conflict severity and type discrimination models using the SVM and on the basis of conflict time positioning are proposed. And the design of the SVM’s input and output variables, the selection of the SVM’s kernel function and the selection of the SVM’s penalty factors are finished. The real vehicle data are discriminated in the conflict time positioning, the conflict severity and the conflict type by proposed methods. The main evaluation indexes of the four kinds of supervised learning models including the producer’s accuracy, the user’s accuracy, the overall accuracy and the Kappa coefficient are selected to evaluate the results processed by the proposed methods, and the evaluated results show that the proposed method can discriminate the period, the severity, and the type of road-traffic-conflict.