众源车辆轨迹加油停留行为探测与加油站点提取

Refueling Stop Activity Detection and Gas Station Extraction Using Crowdsourcing Vehicle Trajectory Data

针对当前加油站点信息采集成本高、更新周期长等问题,提出了运用车辆轨迹数据提取加油站点的方法。首先,从轨迹运动特征、几何模式等方面分析个体和群体加油行为轨迹特征。其次,基于Stop/Move模型,集成加油轨迹运动特征提出速度序列线性聚类算法提取加油停留轨迹。最后,运用Delaunay三角网层次聚类加油停留轨迹线,综合运用傅里叶形状识别、语义约束等方法识别、提取加油站点。运用北京市7d的出租车轨迹数据进行试验分析,共提取482个加油站,正确率为93.1%,且位置精度高。

In view of the deficiencies of current surveying methods of gas station,an approach is proposed to extract gas station from vehicle traces.Firstly,the spatial-temporal characteristics of individual and collective refueling behavior of trajectory is analyzed from aspects of movement features and geometric patterns.Secondly,based on Stop/Move model,the velocity sequence linear clustering algorithm is proposed to extract refueling stop tracks.Finally,using the methods including Delaunay triangulation,Fourier shape recognition and semantic constraints to identify and extract gas station.An experiment using 7 days taxi GPS traces in Beijing verified the novel method.The experimental results of 482 gas stations are extracted and the correct rate achieves to 93.1%.