一种基于轨迹数据的红绿灯位置检测方法

A Traffic Light Position Detection Method Based on Trajectory Data

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。

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 response to the problems of low accuracy and incomplete coverage area of current traffic light position detection methods, we propose 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 applications 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 positions 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 gradient boosting decision tree(GBDT) algorithm is used for classification. Finally, the positive samples of candidate positions 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 experimental 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 acquisition and update of the city's large-scale traffic light position information.