Research On Driver Distraction Recognition And Risk Compensation Method In Mixed Connected Environment

Modern automobiles are developing in the direction of electrification,intelligence,connectivity,and sharing.Researches related to intelligent connected vehicles have received extensive attention from scholars,and considerable achievements have been applied.In the future for a longer period,mixed connected traffic is bound to become normal,where manual driving vehicles,connected manual vehicles,automated vehicles,and connected automated vehicles will simultaneously share the road.Drivers face the complex and changeable mixed connected environment,which increases the characteristics of more randomness and uncertainty,and the research on distraction identification and risk compensation methods in mixed connected environment has become particularly important.Due to the complexity,interactivity and dynamics of the mixed connected environment may change the driver’s distraction characteristics and impact mechanism.At present,the distraction research methods based on traditional environment may not be effectively applicable to the mixed connected environment.Only by clarifying the impact mechanism of driver distraction in the mixed connected environment,quantifying the relationship between distraction and risk,and systematically solving the key problems of driver behavior monitoring can we reduce various traffic accidents caused by human factors in a more scientific,reasonable and effective manner.Therefore,this paper explores the impact mechanism of different types of distractions on driving safety in traditional and mixed connected environments,distraction recognition models,accident probability prediction model and distraction compensation models are established for different traffic environments and distraction types.The specific research contents are as follows:(1)In order to solve the problem that the recognition of behavioral characteristics in mixed connected environment are still unclear,a cognitive distraction recognition model based on ensemble learning algorithm was proposed,the problem of distraction monitoring demand in mixed connected environment is solved.Firstly,the differences in cognitive distraction between the two environments were compared and analyzed.The study found that the mean of saccade speed,the standard deviation of the horizontal fixation angle,the mean of pupil diameter,and the standard deviation of the vertical fixation angle had significant changes.the change of traffic environment had a certain impact on cognitive distraction;Secondly,based on the statistical analysis of driving behavior and visual behavior,a cognitive distraction recognition index system was established respectively.Finally,the learning advantage of the ensemble learning algorithm that can observe the data of different spaces and different structures to correct the bias of the base classifier was used.The Support Vector Machine(SVM),Extreme Gradient Boosting(XGBoost),Long Short-term Memory(LSTM),Attention Mechanism-Bidirectional Long Short-Term Memory(AT-Bi-LSTM)and other superior heterogeneous base classifiers were integrated,and the cognitive distraction recognition model based on Stacking ensemble learning algorithm was constructed.The experimental results show that the proposed model is superior to the traditional algorithm,and the recognition accuracy is 96.04% and 97.83% in traditional and mixed connected environments,respectively.Compared with previous studies,the model proposed in this paper considers the uncertainty caused by traffic environment differences more comprehensively,which enhances the generalization ability and applicability of the model.(2)In order to solve the problem that similar visual-operational distracted driving behaviors are easy to be confused,a cascaded network distraction recognition model considering contextual and temporal relevance is proposed,which solves the problem of difficulty in recognizing similar distracted driving behaviors.Firstly,the State Farm distraction dataset and the self-constructed distraction dataset were used as the research basis,and make time series adjustments to it,and use image enhancement technology to expand the sample size,and the time series distraction image dataset was constructed;Secondly,the backbone network is compared and analyzed,and the lightweight network Mobile Net V3 is selected as the frontend network for convolution extraction of distracting image spatial features.Considering that the recurrent neural network has the advantages of memory and parameter sharing,the Gate Recurrent Unit(GRU)recurrent network is selected as the back-end network for mining contextual time series distraction feature information.Finally,a feature linear fusion strategy is proposed,which realizes the fusion of the front-end depth spatial feature information and the residual of the GRU network output,and makes up for the missing important distracted feature information while connecting the temporal semantics.The recognition accuracy reached 90.92%on the improved State Farm dataset and 89.25% on the self-constructed dataset.The results showed that this model can deeply analyze similar distraction features,and can make full use of spatiotemporal feature information to achieve good identification of distracting behaviors with high similarity.In addition,the model has good robustness and strong applicability to different data sets and easily confusing distracted driving behaviors.(3)In order to solve the problem that the uncertainty of drivers’ distraction risk perception and response characteristics under conflict events,a braking response time model and a probability prediction model of distraction accident were proposed,which solves the problem of unclear relationship between distraction and risk quantification caused by differences in traffic environment.Firstly,the braking reaction time model was constructed based on the generalized linear mixed model,and the potential factors affecting the braking reaction time of different types of distraction in the two environments and variation rules were analyzed.Secondly,a prediction model of accident probability of distracted driving was constructed by using generalized estimating equation,and the impact mechanism and difference of different types of distraction on accident probability in the two environments were explored,which can predict the accident probability under the conflict event.Then,it is verified that when distracted,drivers will take measures to compensate for the risk of distraction by increasing time headway.Finally,a theoretical distraction compensation model is constructed based on the probability prediction theory of distraction,The theoretical compensation under different types of distraction is analyzed and reasoned out,and the effectiveness of theoretical distraction compensation is verified by using actual accident samples.It provides an effective risk avoidance method for dealing with conflict events in traditional and mixed connected environments.