Two-Stage Traffic Clustering Based on HNSW

Traffic flow clustering is a common task to analyze urban traffic using GPS data of urban vehicles. Existing density-based traffic flow clustering methods generally have two important problems, that is to not consider the characteristics of urban roads and not handle well different sizes of urban areas. In this paper, we propose a novel method, called TSST-HDBC (Two-Stage Spatial-Temporal Hierarchical Density-based Clustering), which solves the above problems using a two-stage clustering approach. In the first stage, the characteristics of the input trajectory are considered as a whole in terms of time and space, and an appropriate similarity function is selected. Then, the HNSW (Hierarchical Navigable Small World) structure is used to set different search radii at each layer, and preliminary clustering results are obtained, which are recorded as sub-clusters. In the second stage, the sub-clusters are re-clustered, and the similarity measurement function is applied according to the road characteristics to obtain the final clustering result. Experimental results show that the proposed TSST-HDBC method can effectively solve two problems and improve the accuracy of traffic clustering in urban arears.