A dynamic two-dimensional (D2D) weight-based map-matching algorithm
A dynamic two-dimensional (D2D) weight-based map-matching algorithm
复制标题
DOI：
10.1016/j.trc.2018.12.009
复制
发表时间：
2019-01-01
影响因子：
8.3
通讯作者：
Quddus, Mohammed A.
中科院分区：
文献类型：
--
作者：
Sharath, M. N.;Velaga, Nagendra R.;Quddus, Mohammed A.
现有的地图匹配（MM）算法主要定位定位固定沿着道路的中心线，并在很大程度上忽略了道路宽度作为输入。因此，车辆车道级定位，这是必不可少的严格的智能交通系统（ITS）的应用，似乎很难完成，特别是与定位数据从低成本的GPS传感器。本文旨在通过开发一种新的动态二维（D2 D）加权MM算法，结合动态权重系数和道路宽度，以解决这一限制。为了实现车辆车道级定位，道路段被虚拟地表示为关于道路中心线的均匀网格的矩阵。这些网格，然后使用地图匹配定位固定，而不是在传统的MM算法进行道路中心线上的匹配。在这个开发的算法中，在路段上的车辆位置识别是基于总权重分数，这是四个不同权重的函数：（i）接近度，（ii）运动学，（iii）转弯意图预测，和（iv）连接性。采用自适应回归的方法，利用不同的网络复杂度和定位质量参数，将相对重要性分配给不同的权值。为了证明所开发算法的可移植性，使用在英国诺丁汉收集的5，830个GPS定位点以及在印度孟买和浦那收集的7，414个GPS定位点进行了测试。开发的算法，使用独立的GPS定位，确定正确的链接96.1%（诺丁汉数据）和98.4%（孟买-浦那数据）的时间。在正确的车道识别方面，该算法被发现提供了84%（诺丁汉）和79%（孟买-浦那）的固定由独立的GPS获得的准确匹配。使用本研究中采用的相同方法，如果利用来自附加传感器（例如陀螺仪）的定位数据，则可以进一步提高车道识别的准确性。ITS行业和车辆制造商可以将这种D2 D地图匹配算法用于责任关键和车载信息系统和服务，如高级驾驶员辅助系统（ADAS）。
Existing map-Matching (MM) algorithms primarily localize positioning fixes along the centerline of a road and have largely ignored road width as an input. Consequently, vehicle lane-level localization, which is essential for stringent Intelligent Transport System (ITS) applications, seems difficult to accomplish, especially with the positioning data from low-cost GPS sensors. This paper aims to address this limitation by developing a new dynamic two-dimensional (D2D) weight-based MM algorithm incorporating dynamic weight coefficients and road width. To enable vehicle lane-level localization, a road segment is virtually expressed as a matrix of homogeneous grids with reference to a road centerline. These grids are then used to map-match positioning fixes as opposed to matching on a road centerline as carried out in traditional MM algorithms. In this developed algorithm, vehicle location identification on a road segment is based on the total weight score which is a function of four different weights: (i) proximity, (ii) kinematic, (iii) turn-intent prediction, and (iv) connectivity. Different parameters representing network complexity and positioning quality are used to assign the relative importance to different weight scores by employing an adaptive regression method. To demonstrate the transferability of the developed algorithm, it was tested by using 5,830 GPS positioning points collected in Nottingham, UK and 7,414 GPS positioning points collected in Mumbai and Pune, India. The developed algorithm, using stand-alone GPS position fixes, identifies the correct links 96.1% (for the Nottingham data) and 98.4% (for the Mumbai-Pune data) of the time. In terms of the correct lane identification, the algorithm was found to provide the accurate matching for 84% (Nottingham) and 79% (Mumbai-Pune) of the fixes obtained by stand-alone GPS. Using the same methodology adopted in this study, the accuracy of the lane identification could further be enhanced if the localization data from additional sensors (e.g. gyroscope) are utilized. ITS industry and vehicle manufacturers can implement this D2D map-matching algorithm for liability critical and in-vehicle information systems and services such as advanced driver assistant systems (ADAS).