| With the rapid development of the social network in the Internet,the trajectory mobile network in the physical world has received increasing attention.It has become the focus of trajectory detection to mine the similarity of trajectories from users in the same time segment and get accurate trajectory features.However,in real life,the trajectory data generated by user activities are relatively dense,and the number of trajectories increases in time segments.The display of trajectory geographic information on the map is unregulated,and the trajectory data in daily activities has the problem of user behavioral crossover.Therefore,it is difficult to analyze the behavior of the trajectory dataset effectively.The trajectory data mainly includes five characteristics of user coding,t ime spot,longitude,latitude and the height of the geographical location.For the behavior analysis of trajectory data with large density and high degree of the crossover,considering the negligible influence of height difference,this paper proposes single-source spatial feature model based on sparse sub-segment,which transforms the trajectory data into time series defined by longitude and latitude.The proposed algorithm firstly measure spatial similarity between the trajectories from users,then performs sparse analysis on the similarity,and finally obtains global optimal subset clustering.The algorithm framework of this paper is mainly as follows.(1)The time series mining-dynamic time rectification principle is adopted to normalize the minimum distance between trajectories,which is used to define the spatial similarity of the trajectory matrix.Moreover,the row-sparsity regularization is imposed on the feature matrix to enhance the performance of the sparse subsets.(2)By modeling the latitude and longitude of each trajectory accurately,the trajectories with similarity characteristics are obtained to represent the information of the dataset,which maintains the minimum distance metric to the surrounding trajectory.(3)The labels of the feature trajectories are propagated onto other trajectories with similarity relations.Different label groups represent different trajectory groups,each of which stands for a relatively independent community.The connection between the trajectories within the community is strong and the association between communities is weak,which ensures the characteristics of cluster analysis.Experiments were carried out on the driving trajectory datasets of all taxis in Beijing in one week.The clustering number and cluster distribution of the trajectory dataset are analyzed under the different constraint parameters of the objective function.Meanwhile,the algorithm is compared with the dense sub-graph detection method.The comparison of the clustering result and the running time of the algorithm proves the effectiveness and efficiency of our method. |
