Abstract:

With the widespread use of technologies such as the Internet of Things， V2X and smart cities， a large amount of vehicle trajectory data is recorded and retained. The trajectory data can be used to extract relevant information， such as calculating the optimal path， detecting abnormal driving behavior， monitoring urban traffic flow and predicting the next location of vehicles， etc. for which trajectory clustering is one of the key technologies. Density Peaks Clustering （DPC） is a simple and effective density-based clustering algorithm， but the definition of local density in the algorithm does not fully consider the influence of density difference when the density of data samples is unevenly distributed， nor does it have a similarity measure suitable for vehicle driving track. In addition， the algorithm is not effective when encountering relatively high dimensional data. By introducing in k-Nearest Neighbor （KNN） and Principal Component Analysis （PCA） and improving similarity measurement， this paper proposes a density peak clustering algorithm suitable for vehicle running trajectory. Firstly， PCA is used to preprocess high-dimensional data. Then the local density is redefined using the K-nearest neighbor idea. Finally， the influence of Euclidean distance on the allocation strategy is abandoned by redefining the distance function between tracks. The feasibility of the algorithm is proved by experiments on synthetic data sets. At the same time， the effectiveness of the algorithm is also verified by the actual vehicle trajectory data.

Key words: density peaks xlustering, trajectory data, k-nearest neighbor, principal component analysis, measure of similarity