一种基于统计推断模型的车辆GNSS轨迹滤波方法

发布时间：2018-04-30 22:02

本文选题：信息熵 + 最大似然分类　；参考：《地理与地理信息科学》2017年05期

【摘要】：车辆GNSS轨迹是由离散定位点连成的折线段,以往轨迹滤波工作注重离群点及其引起的轨迹线段异常,而忽视了由于采样间隔长而导致的正常点连线严重偏离道路的现象。为解决此问题,该文提出一种基于统计推断的轨迹滤波方法,不依赖路网等辅助数据,在去除离群点的基础上,识别出上述低质量线段。首先采用基于信息熵的最大似然分类法确定线段端点类型,即该端点是位于路口还是道路上;然后再构建规则推断模型,根据线段两端点的空间组合方式,检测出线段是否异常;最后利用北京市2012年11月出租车数据进行验证,采样间隔为50~65s。结果表明,离群点导致的异常线段占32.21%,低采样率导致的亦占6.23%,滤波效果整体良好。
[Abstract]:The GNSS trajectory of vehicle is a broken line segment composed of discrete positioning points. In the past, trajectory filtering focused on outliers and the abnormal track segments, but neglected the phenomenon that the line of normal points deviated seriously from the road due to the long sampling interval. In order to solve this problem, a trajectory filtering method based on statistical inference is proposed in this paper, which does not rely on auxiliary data such as road network, and identifies the low quality line segments above on the basis of removing outliers. Firstly, the maximum likelihood classification method based on information entropy is used to determine the type of segment endpoint, that is, whether the endpoint is located at an intersection or on a road, and then a rule inference model is constructed, which is based on the spatial combination of the points at both ends of the line segment. Whether the line segment is abnormal or not is detected. Finally, the taxi data of Beijing in November 2012 are used to verify it. The sampling interval is 50 ~ 65s. The results show that the abnormal line segment caused by outliers is 32.21, and that caused by low sampling rate is 6.23. The filtering effect is good.
【作者单位】：中国科学院大学;中国科学院光电研究院;中国科学院地理科学与资源研究所;
【基金】：国家重点研发计划(2017YFB0503702) 国家自然科学基金项目(41601486) 中国科学院知识创新工程重要方向性资助项目(KZCX2-EW-QN605)
【分类号】：TN967.1;U495

【相似文献】