Testing Scenario Extraction And Simulation Of Vehicle-to-vehicle Cooperation Based On Traffic Accident Reports

Data-driven scenario construction technology is currently one of the popular methods in the field of ICV(Intelligent and Connected Vehicles)testing.At present,researchers mostly use structured natural driving data for scenario research,while ignoring the unstructured traffic accident report with abundant data.And manual analysis is often indispensable in the process of scenario construction,resulting in a long scenario construction cycle and low efficiency.This paper takes the research status of ICV testing-scenario generation technology as the background,and aims at the problems of low utilization and low degree of automation in the current traffic accident data in the test scenario construction.Scenario-generation methods are proposed which combine domain ontology and natural language processing.This method realizes the automatic extracting scenario information from the accident report.At the same time,aiming at the problems of the current test platform and evaluation system is not perfect,the test scale is small and so on.This article build a simulation framework for testing scenario by using SUMO simulator.This framework realizes the testing and evaluation of V2 V application.This article is based on the National key research and development program(2018YFB1600800).The main research contents are as follows:(1)Ontology construction in the field of traffic accident.The primary task of scenario extraction is to construct the ontology of traffic accident.In order to achieve this goal,on the basis of determining the scope of ontology application,this article combines the knowledge system contained in the accident report,draws on the sevenstep method and protégé tool construction process,the six generic structures of accident domain ontology are designed(including vehicle,obstacle,action,environment,road and accident),17 kinds of object attributes,7 kinds of data attributes and some instances.Finally,the ontology is constructed and saved by protégé tool.(2)Design and verification of accident scenario information extraction framework.This paper proposes an information extraction framework combines ontology and relation extraction,among them,the accident ontology is used to extract the entities in the text,and relation extraction is used to identify the correlation between entities,thereby improving the accuracy of scenario information extraction.Firstly,78 sets of accident domain dictionary templates are constructed to process the text.Then,entity relation is extracted by using dependency relation technology,and accident information extraction algorithm and scenario reasoning rules are designed by combining ontology knowledge.Finally,the test is carried out on 100 accident data.The results show that the integrity and accuracy of information extraction are 89% and 98.33%,respectively.The average time of processing a single report is 297 ms,which meets the requirement of automatic extraction.(3)The design and verification of virtual simulation based on the SUMO simulator.This paper proposes a virtual scenario simulation testing framework based on SUMO platform.The framework considers how to insert the accident scenario into the driving environment of the test vehicle.Through the scenario insertion and V2 V application module,it is possible to complete the scenario test task on the sumo simulator.Combined with TTC,following distance,speed and longitudinal acceleration,the overall evaluation of the simulation test is carried out.On the basis of the framework,experiment 1 selects three typical scenarios for 30 groups of tests to verify the effectiveness of the scenarios.Experiment 2 designs two FCW(forward collision warning)applications with the same function and different complexity,and tests their performance.The experimental results show that: Experiment 1 shows that the accident scenario injection can reduce the safety and comfort index of the test vehicle,which proves that the test scenario has the effectiveness of the test network connection function.In Experiment 2,by comparing the differences between different V2 V models in terms of safety and comfort,it is concluded that the FCW model with high complexity has better performance in terms of safety and comfort.