首页|一种基于轨迹数据的红绿灯位置检测方法

一种基于轨迹数据的红绿灯位置检测方法

扫码查看
原文链接
  • NETL
  • NSTL
  • 万方数据
红绿灯位置是道路上行人和车辆的交会点,极大影响着道路结构和交通运行,在城市路网中起着重要的枢纽作用.针对目前红绿灯位置检测方法准确率不够高、覆盖面区域不完整等问题,提出了一种基于轨迹数据的交通灯位置检测方法.该方法基于聚类-合并-分类-合并的四级模型,首先从清理过的轨迹数据中提取隐含的车辆行驶特征,再采用具有噪声的基于密度的聚类(density-based spatial clustering of applications with noise,DBSCAN)方法得到转向和停驻两类聚类中心,对这两类聚类中心进行合并,获得红绿灯位置的候选位置;根据候选位置一定范围内的轨迹点提取该区域的车流行驶特征,然后采用梯度提升决策树(gradient boosting decision tree,GBDT)算法进行分类,最后将候选位置的正样本融合,以检测红绿灯位置.采用成都市浮动车GPS轨迹数据进行实验,检测结果的F1 分数为0.947,效果优于常规的机器学习方法.实验结果表明,基于GPS轨迹数据,采用提出的四层模型能有效检测出红绿灯的位置,该模型可被用于城市大范围红绿灯位置信息的快速获取和更新.
A Traffic Light Position Detection Method Based on Trajectory Data
The position of traffic lights is the intersection of pedestrians and vehicles on the road,which greatly affects the structure of the road and the operation of traffic,and it plays an important pivotal role in the urban road networks.In re-sponse to the problems of low accuracy and incomplete coverage area of current traffic light position detection methods,we pro-pose a trajectory data-based traffic light position detection method.Based on the four-level model of clustering,merging,classifying and merging,this method extracts the hidden driving characteristics of vehicles from the cleaned trajectory data,and then uses density-based spatial clustering of applica-tions with noise(DBSCAN)method to get the two types of cluster centers(steering and stationary).Then these two types of cluster centers are merged to obtain the candidate po-sitions of the traffic light.According to the trajectory points within a certain range of the candidate positions,the vehicle flow characteristics in this region are extracted.And then gra-dient boosting decision tree(GBDT)algorithm is used for classification.Finally,the positive samples of candidate posi-tions are fused to detect traffic light positions.We use the floating car GPS trajectory data in Chengdu for experiments.The F1 score of the prediction result is 0.947,which is better than that of conventional machine learning methods.The ex-perimental results show that the proposed four-level model can effectively detect the position of the traffic light according to the GPS trajectory data,and it can be used for the rapid ac-quisition and update of the city's large-scale traffic light posi-tion information.

urban trafficfloating carroad networkspatio-temporal characteristicstraffic light position detectionGPS trajectorygradient boosting decision tree(GBDT)DBSCAN

赵肄江、方辰昱、廖祝华

展开 >

湖南科技大学计算机科学与工程学院,湖南 湘潭,411201

湖南科技大学服务计算与软件服务新技术湖南省重点实验室,湖南 湘潭,411201

城市交通 浮动车 道路路网 时空特征 红绿灯位置检测 GPS轨迹 梯度提升决策树(gradient boosting decision tree,GBDT) DBSCAN

国家自然科学基金湖南省教育厅科学研究重点项目湖南省自然科学基金

4187132019A1722021JJ30276

2024

10.14188/j.2095-6045.2022657

测绘地理信息
武汉大学

测绘地理信息

CSTPCD
影响因子：0.563
ISSN：1007-3817
年,卷(期)：2024.49(2)
  • 22