基于约束高斯混合模型的车道信息获取

针对现有车道级道路信息获取方法大多存在数据采集成本高、更新周期长、数据处理难度大等缺点，提出了一种基于浮动车数据（floating car data，FCD）的城市车道数量信息快速获取方法。首先根据浮动车数据的空间分布特征，利用Delaunay三角网方法对数据进行优选，通过探测优选后浮动车数据覆盖的宽度间接得到道路宽度；然后将一部分已知车道数量及浮动车数据覆盖宽度的路段作为训练样本，分析其车道数量和浮动车数据覆盖宽度之间的关系构建基本分类器；最后按照待测路段的浮动车数据分布宽度查找基本分类器，获取待测路段可能存在的若干个车道数量类型候选值，并利用约束高斯混合模型对最终车道数量类型进行确认。实验结果表明，该方法实现了从低精度浮动车数据中快速获取车道数量信息，提取精度达到了82.3%。

Traffic Lane Number Extraction Based on the Constrained Gaussian Mixture Model

The present methods are expensive and time-consuming that can't keep up with changes in the roads due to construction or disasters. Instead of this, an attractive alternative is proposed to acquire lane numbers from Floating Car Data collected by taxis. Firstly, the Delaunay triangulation network method is used to optimize the raw GPS trajectories by considering the spatial distribution characteristics of floating car data, and then the road width is set by detecting the trajectories covered width. The primary classifier is established from training data that shows the relation between lane numbers and road width, and some candidate lane numbers can be inferred by detecting the trajectory covered width with the primary classifier. The final lane numbers are determined by fitting GPS traces using the Constrained Gaussian mixture model. Experiments results show the method proposed in this paper can acquire lane numbers from low-quality floating car data and the overall accuracy is 82.3%.