关键词: 轨迹数据, 道路交叉口, 密度峰值聚类, 低频率, 数据融合, 数学形态学

Abstract:

Taxi GPS trajectory data are of low acquisition cost, short cycle, large coverage, large-scale, and real-time. Moreover, taxi trajectory data contain a large amount of driving record information for extracting digital road information. Thus, taxi GPS trajectory data are suitable for obtaining and rapidly updating the information of large-scale urban traffic road networks. The extraction of urban road network intersections based on GPS trajectory data is currently a research hotspot. However, most of existing methods, which are applicable to the high-frequency GPS data, are difficult to adapt to taxi trajectories with low sampling frequency, low positioning accuracy, many noise points, and uneven data distribution. Therefore, existing methods are not readily applied to extract the intersections of suburb areas where taxi trajectory data are sparse or low-frequency. To extract road intersection information as accurately and comprehensively as possible, this paper proposed an integrated methodology to identify the intersections of urban road networks based on dense and sparse trajectory data. In this paper, the density peak clustering method was adopted in the vector space. Meanwhile, the mathematical morphology processing method was adopted in the grid space, where multiple resolution images were generated in the trajectory data rasterization stage. The extraction results were finally fused to achieve the purpose of extracting the road intersections of suburb areas with low traffic (i.e., sparse sampled data). Further, a fusion mechanism was designed to detect these intersections by fusing multiple results in both spaces. Finally, this paper used principal component analysis to determine the authenticity of the intersections, which was used to identify real intersections and remove pseudo intersections that were incorrectly extracted. In so doing, we obtained the urban road intersections based on the low-frequency taxi trajectory. Compared with existing methods, this method extracted more intersections and showed considerable consistency with remote sensing imagery. Besides, the accuracy evaluation shows that the extraction accuracy was 92.23%, the recall rate was 77.26% and the F-value was 84.08%. Our findings suggest that the proposed methodology can ensure the integrity and accuracy of urban road network intersections and be applied in intelligent transportation systems.

Key words: trajectory data, road intersections, density peak clustering, low-frequency, data fusion, mathematical morphology