基于浮动车数据的路口探测方法  

Detection of Road Intersections Using Floating Car Data

作　　者：谭祥爽 王静 宋现锋[1,2] 许长辉[3] 汪超亮[2] 

机构地区：[1]中国科学院大学,北京100049 [2]中国科学院光电研究院,北京100094 [3]中国测绘科学研究院,北京100073

出　　处：《地理与地理信息科学》2015年第5期34-38,F0003,共6页Geography and Geo-Information Science

基　　金：国家科技重大专项(2011ZX05039-004);中国科学院知识创新工程项目(KZCX2-EW-QN605);国家自然科学基金项目(40771167)

摘　　要：路口及其通行规则探测是当前路网自动化提取工作中的难点问题,该文提出了一种简单高效的从浮动车数据中探测路口与转向规则的方法。假设车辆航迹线转弯密集区为路口,首先通过对转弯曲线空间聚类分析获得路口中心,再利用同心圆面积递增法估算出路口覆盖范围,最后根据路口范围内航迹线行驶方向分布模式,识别出通行规则。该方法的特点是将路口视为具有中心和半径的区域,而不是简单的点。将该方法应用于淮北市城区出租车采集的浮动车数据,同人工解译结果相比,探测路口的正确识别率约为91.6%,路口位置与解译结果平均距离为3.57m,表明方法具有较高的实用价值。The detection of intersections is a hotspot and difficulty issue in the automated extraction of road network from floating car data（vehicle GNSS trajectories）.This paper presents a simple but efficient approach for detecting intersections and associated turning rules.Assuming intersections are located in the areas where exists local maxima of dense turning curves of vehicles,the locations of intersection are first detected by the MeanShift algorithm,the sizes of intersections are then estimated by the proposed concentric area increment method,and the turning rules are finally determined by classifying the directions of both entrance and exit of those approaching trajectories.The advantage of this work is that an intersection is treated as a compact area in which traffic rules are used to schedule vehicles approaching,not a simplified point.This approach was tested using taxi trajectories collected in Huaibei City,China.In a comparison of visually interpreted results,more than 91.6% of intersection could be correctly recognized and the average distance between the result and the truth is 3.57 m.This shows a big potential advantage of applying the approach in generating routable maps.

关 键 词：浮动车数据 路口识别 转向规则 空间聚类 MEANSHIFT算法 

分 类 号：P283.7[天文地球—地图制图学与地理信息工程]