车辆轨迹数据提取道路交叉口特征的决策树模型

The vehicle trajectory data from multiple sources implies the latest road distribution information. Research on using trajectory data to extract road features is beneficial for the rapid establishment and update of basic road network data. The road network is composed of intersections and road lines connecting intersections, among which the identification of intersection features is the key to the generation of the entire road network. Due to the lack of a refined intersection identification model, the road network generated from trajectory data is prone to phenomena such as missed intersections and structural distortion. To address this problem, this paper proposes a method for extracting road intersections using trajectory data. Firstly, the changes in the geometric shapes and implicit dynamic characteristics of vehicle movement trajectories in intersection and non-intersection areas are analyzed. Then, a trajectory segment classification model is constructed using the decision tree method, and an extraction method for lane-changing trajectory segments in intersection areas is established in combination with a moving window-type trajectory line segmentation model. Finally, the trajectory segments in the intersection area are clustered according to the Hausdorff distance, and the center line is extracted to obtain a complete road intersection structure. Real vehicle trajectory lines are used as test data to verify the effectiveness of the method proposed in this paper.

众源车辆轨迹数据隐含最新的道路分布信息,研究利用轨迹数据提取道路特征有益于基础路网数据的快速建库与更新。道路网由交叉口和连接交叉口的道路线构成,其中交叉口特征识别是整个道路网生成的关键。由于缺乏精细的交叉口识别模型,轨迹数据生成的道路网容易出现路口遗漏、结构失真等现象。针对这一问题,本文提出一种利用轨迹数据提取道路交叉口的方法。首先,分析车辆在交叉口与非交叉口区域移动轨迹几何形态及隐含动力学特征的变化情形;然后,利用决策树方法构建轨迹片段分类模型,并结合移动开窗式的轨迹线剖分模型建立交叉口区域变道轨迹片段提取方法;最后,依据Hausdorff距离对交叉口区域轨迹片段进行聚类,并提取中心线获得完整的道路交叉口结构。采用真实的车辆轨迹线作为测试数据,验证了本文提出方法的有效性。