The Identification And Analysis Of High-risk Event Based On Naturalistic Driving Study

Traffic high-risk events are highly likely to cause traffic congestion,and even provoke road traffic accidents,aggressive driving behavior and other malignant events.As a general concept of traffic safety,"high-risk" indicates the degree of risk potential,i.e.,prone to road traffic accidents.High-risk events essentially consist of two dimensions.One is at the level of traffic flow,due to individual interactions between vehicles caused by the disruption of traffic flow,which leads to a significant increase in the probability of traffic accidents,called traffic operation high-risk events.The other is at the level of driving behavior,due to the driver’s own state(such as distraction,fatigue and other driving behavior),or due to the interaction between the vehicles triggered by anger and aggressive driving behavior,which in turn leads to a significant increase in the risk of accidents,called driving behavior high risk events.Such two dimensions of high-risk events are both characterized by low occurrence-probability and dynamics.And the analysis,identification and early warning of high-risk events have been the focus and difficulty of research in the field of traffic safety.Therefore,this thesis is based on the naturalistic driving environment,presents modeling,identification and characteristic analysis for the two typical high-risk events,and tries to construct a generalized high-risk event identification and analysis method.Both traffic operation high-risk events and driving behavior high-risk events face the difficulty of characterizing indicators to truly link micro-behavior and traffic flow states.For the traffic operation high-risk events,it focuses on the interaction behavior between vehicles in the traffic flow and the analysis of the overall traffic flow status,but identifying them from the micro level of vehicle operation is more adapted to the current reality of progressively abundant vehicle trajectory data,and helps to improve the accuracy of identification and early warning,as well as improve the analysis of influencing factors.While for the high-risk events in driving behavior,although it is a specific manifestation of individual vehicle and micro driving behavior,its triggers and behavior manifestation are inextricably linked to the overall traffic flow environment.Therefore,the key to constructing a generalized high-risk event identification and analysis method is to link the two with a unified scale that can portray both individual microscopic behavior and characterize the overall traffic flow state,and to quantify the degree of abnormality of individual microscopic behavior and characterize the traffic flow state from the perspective of microscopic interaction behavior.This research is precisely from the above research needs,introduce and define the concept of traffic entropy from the basic concept of entropy and the evolution of information entropy.Through traffic entropy to quantify the microscopic driving behavior of individual vehicles and the interaction behavior between vehicles,and then characterize the state and disorder of traffic flow,which laying the theoretical foundation for the identification and analysis of high-risk events in traffic operation and driving behavior.In the application of traffic operation high-risk events,considering the reality that the connected and autonomous vehicles(CAVs)are gradually penetrating into the transportation system and the low-ratio of CAVs scenario will exist for a long time,we propose an early warning method for high-risk events of traffic operation under low ratio of CAVs.Specifically,the microscopic driving behavior parameters of individual vehicles are quantified as traffic entropy to characterize the traffic flow state.And then the traffic entropy is used as the input parameter of the LSTM model to establish the early warning model of high-risk events.The High D Dataset from German highways was utilized for the empirical analyses.In order to compare the application results under CAVs environment,an autonomous-vehicles scenario and a connected-vehicles scenario were set for the high-risk events and non-risk events extracted from the High D Dataset.and the effectiveness of the warning of high-risk events under different vehicle permeability was compared.Results show that,the false alarm and missed alarm rates of early warning model with traffic entropy parameters are both reduced.Taking the low-ratio CAVs of 10% as an example,the false alarm and missed alarm rates reduced from 6.18% and 11.47% to 1.95% and 3.12%,respectively.At the same time,the false alarm and missed alarm rates are only 2.28% and 3.82% under the prediction environment(2-3 seconds advance warning).In the application of high-risk events in driving behavior,we focus on road rage driving behaviors which are typically characterized by subjective intent and random aggression,are strongly associated with personal heterogeneity of human drivers,and have rarely been studied for identification and analysis starting from vehicle kinematic data.In response to the vague and ill-defined problem,the definition of road rage driving behavior was first clarified,and the process of road rage events was analyzed to construct a complete road rage event chain and three elements(trigger,driver anger state and road rage manifestation).Taking advantage of traffic entropy in quantifying microscopic driving behaviors,a method is proposed to initially identify suspected road rage driving behaviors from naturalistic driving processes using vehicle kinematic data.The method extracts key behavioral characterization parameters based on behavioral features and driving style heterogeneity classification,and uses traffic entropy as a quantitative indicator to filter out suspected road rage driving behaviors from massive naturalistic driving data.Next,computer video recognition technology is used to capture the feature values of key points of the driver’s skeleton from the video images to assist in confirming the driver’s abnormal anger state and behavior.Finally,an road rage trigger scale and a road rage performance behavior scale were established to construct a complete road rage event chain.Based on the massive Chinese naturalistic driving dataset,200 road rage events containing the three elements of the complete road rage event chain were identified from more than 18,000 naturalistic driving trips of 53 drivers.Based on the complete road rage event chain,a two-layer logistic regression model was constructed to analyze the influencing factors of road rage events from the driver and event characteristics layers,respectively.The analysis focused on the characteristics of high-anger drivers,the triggers of road rage events and the manifestation of road rage behaviors.Results show that,(1)Drivers with aggressive carfollowing behavior were less likely to experience road rage events than others.(2)Slow driving is the most important trigger for road rage events,and scenarios with a combination of multiple triggers further increase the probability of road rage events.(3)The relationship between some of the road rage driving behavioral manifestations and their corresponding triggers can be explained by behavioral compensation mechanisms.At the same time,the three elements of the road rage event chain are found all follow the 80/20 rules,i.e.,80% of road rage incidents are composed of/caused by 20% of the key factors.The above findings helps to provide scientific suggestions and guidance on targeted measures to reduce or prevent road rage generation from different perspectives,such as social promotion,driver education,traffic management,and policy design.This thesis tackles the difficult problem of identifying high-risk events in naturalistic driving environments,proposes the concept of traffic entropy from the perspective of individual micro-interaction behaviors,quantifies the degree of abnormality of micro-driving behaviors and the disorder of traffic flow states.And the correlations with accident risk were further analyzed.The methods proposed by this thesis lays down important theoretical and methodological support for the control of high-risk events in natural driving environments.