关键词: 道路交叉口, 轨迹数据, 特征提取, 轨迹栅格化, 深度学习

Abstract: Vehicle trajectory data provides a new opportunity for road network generation, road map update and traffic condition monitoring. Accurately extracting road intersections from trajectory data is a key step to build a refined road network map based on vehicle trajectory data. At present, several scholars have put forward some methods using spatial clustering to identify road intersections based on the detection of turning points and speed change positions in trajectories. However, due to the heterogeneity of track density distribution, noise interference and the issue of clustering parameters setting, the existing methods still have challenges to extract intersections of different sizes and shapes from low-quality trajectory data. Therefore, this paper puts forward a strategy of track rasterization considering the heterogeneity of track density distribution and a road intersection extraction method based on the trajectory transformation, intersection segmentation and location optimization process. Experiments on real-world trajectory data with different sampling frequencies are conducted to evaluate the performance of the proposed method, and results show the effectiveness and superiority of the proposed method over the existing state-of-the art methods.

Key words: road intersection, trajectory data, feature extraction, track rasterization, deep learning