Risk Assessment Of Diversion Area Lane Change Considering Visual Characteristics

Freeway diverging areas is the safety bottleneck of freeway.In order to effectively evaluate the safety service level,this paper uses the D-Lab system to collect and analyze the eye-movement data,and uses the traffic conflict technique to extract trajectory.Last,a risk evaluation model is proposed.The specific research contents are as follows:(1)The experimental scheme of acquiring visual characteristics is proposed.A total of 25 drivers are selected to conduct field trials at the parallel optional-single-lane diverging areas.Then,the eye movement data are obtained and the references are introduced to convert the pixel coordinates into unique two-dimensional coordinate.(2)Based on the interest areas,the characteristics of eye movement are analyzed.This paper uses the binary variable analysis to divide the gaze distribution into two categories.The gaze distribution of senior drivers is divided into seven regions using the affinity propagation clustering algorithm which the front window is divided by radiation.The number of clusters is established by adjusting the damping coefficient ? and the deflection parameter p.Then,the difference analysis between driving behaviors is realized and eleven differences are selected.(3)The decision model of lane chang is constructed using the index system.The principal component analysis is proposed to reduce dimension.Based on the non-linear support vector machine,a classification algorithm is presented which distinguish the driving behaviors by probability.The final decision-making is determined by comparing four kinds of kernel functions.The results show that the accuracy and sensitivity of radial basis function are 91.67%and 90.21%,which is suitable for small sample size and low dimension.(4)Based on the predicted trajectory,this paper proposes a method to identify conflict severity.The trajectories are predicted by neural network algorithm.Then,the quantitative indicator J is introduced to analyze the conflict severity using probability.Meanwhile,a collision probability algorithm for fusion lane change decision is discussed,and the risk evaluation model at diverging areas considering visual characteristics is constructed.The results show that the identification accuracy of conflict severity is improved using J which takes the danger-averting action into consideration,the accuracy and sensitivity of the fusion model are 85.71%and 92.75%,which is closer to the actual.