基于K均值聚类挖掘的跟车测试场景研究

Research on Car-Following Test Scenarios Based on K-Means Clustering Mining

为满足自动导航辅助驾驶系统(Navigate On Pilot,NOP)跟车巡航部分的研发需求,基于K均值聚类算法,对跟车场景进行聚类挖掘研究。首先进行实车道路试验,从自然驾驶数据中筛选出1010例符合提取需求跟车场景片段,并在此基础上对跟车场景进行场景解构,提取跟车场景的交通环境要素和场景主体要素。由于跟车场景的场景主体要素数量繁多,采用PCA算法对场景主体要素进行降维,并对筛选的1010例场景片段进行K均值聚类,挖掘到了3种典型的跟车场景,结合交通环境要素及车辆类型等要素,最终获得了大量的测试场景。

To meet the research and development needs of the car-following cruise part of the NOP(Navigate On Pilot),the clustering mining of car-following scenes is studied based on K-Means clustering algorithm.Firstly,through the road test of real vehicles,car-following scene segments of 1010 are selected from the natural driving data,and on this basis,the car-following scenes are deconstructed,and the traffic environmental elements and main elements of the car following scenes are extracted.Due to the large number of main elements in the car-following scenes,PCA(Principal Component Analysis)algorithm is used to reduce the dimension of the main elements for the car-following scenes,and K Means clustering is carried out on the screened scene fragments of 1010,and 3 typical car-following conditions are mined.Combined with traffic environmental elements and vehicle types,1728 test scenarios are finally obtained.