摘要
针对当前网联车车辆与行人交互(以下简称人车交互)测试场景复杂性低、无法充分涵盖真实道路情况下的人车交互场景测试需求问题,提出了一种基于自然驾驶数据的人车交互测试场景构建方法。该方法首次提出了二阶人车交互测试场景的概念,在传统人车交互场景基础上组合前序车辆运动场景,实现交互场景时间域扩充,并通过挖掘自然驾驶数据中人车交互场景及其前序场景的关联性与特征参数,构建更加符合真实情况的复杂人车交互测试场景。首先,在795776条人车交互数据基础上筛选出135起有效人车交互事件及其前序事件,通过统计分析得到典型前序事件——跟车事件。其次,对跟车事件与人车交互事件进行特征挖掘聚类,得到3类跟车场景与4类一阶人车交互场景;将一阶人车交互场景与跟车场景进行拼接组合,通过联合频数统计,从场景拼接组合中提取出6类典型二阶人车交互场景。最后,基于IDM(Intelligence Driving Model)跟车模型与典型AEB(Autonomous Emergency Braking)控制算法,建立了二阶场景下单车环境与网联车环境数值仿真模型,对6类典型二阶人车交互场景生成的测试用例进行评估。结果表明:对于6类二阶场景,网联车环境危险测试用例总量均较单车环境情况有显著下降;对相同一阶场景进行时间域扩充得到的二阶人车交互场景,使得单车环境和网联车环境下的危险测试用例数量都存在着显著差异;提出的二阶人车交互场景构建方法,可以挖掘出更多接近真实道路情况下潜在的危险人车交互测试场景,实现对网联车辆充分、有效的测试。
The most existing test scenarios for the interaction between connected vehicles and pedestrians(hereinafter referred to as V2P interaction)are not sophisticated enough to fully cover the test requirements of V2P interaction scenarios on the real road.A method based on naturalistic driving data was proposed to construct the test scenarios.A new concept,named two-stages test scenario for V2P interaction,was introduced in this method.By combining the classic V2P interaction scenario and the previous scenario of the involved vehicle,a two-stages test scenario expands the time domain of the V2P interaction scenario.Through mining the correlation and characteristic parameters of V2P interaction scenarios and their previous scenarios within naturalistic driving data,sophisticated V2P interaction test scenarios were constructed accordingly.First,based on a total of 795776 records of V2P interaction data,135 effective V2P interaction events and their previous events were selected,and the critical previous events,car-following,were obtained through statistical analysis.Second,feature mining and clustering for the car-following events and the classic V2P interaction events were carried out to obtain 3 types of car-following scenarios and 4 types of one-stage V2P interaction scenarios.The one-stage V2P interaction scenarios and the car-following scenarios were concatenated and combined,based on which 6 critical two-stage scenarios were extracted through joint frequency analysis.Finally,using the car-following model of IDM(Intelligence Driving Model)and a typical AEB(Autonomous Emergency Braking)control algorithm,numerical simulation models were created to evaluate the extracted 6 two-stages V2P interaction scenarios for both,non-connected vehicles environment and connected vehicles environment.The results showed that for all 6 types of two-stage scenarios,the total number of dangerous test cases for connected vehicles is significantly lower than that of non-connected vehicles environment.Moreover,for different two-stage scenarios derived by expanding the time domain of the same one-stage scenario,there is a significant difference in the number of dangerous test cases for both,non-connected vehicles environment and connected vehicles environment.In conclusion,the two-stages V2P interaction scenario construction method proposed in this study can extract more potential dangerous V2P test scenarios close to real road conditions and achieve sufficient and effective testing of connected vehicles.
作者
马峻岩
田叶凡
赵祥模
王卓
柳有权
MA Jun-yan;TIAN Ye-fan;ZHAO Xiang-mo;WANG Zhuo;LIU You-quan(School of Information Engineering,Chang'an University,Xi'an 710064,Shaanxi,China)
出处
《中国公路学报》
EI
CAS
CSCD
北大核心
2022年第3期139-152,共14页
China Journal of Highway and Transport
基金
国家重点研发计划项目(2018YFB1600800).
关键词
交通工程
测试场景
聚类
自然驾驶数据
行人与车辆交互
traffic engineering
test scenario
clustering
naturalistic driving data
interaction between vehicle and pedestrian
