Representative Trajectory Generation Based On Fast Density Peak Clustering

With the popularization of sensors,the rise of mobile Internet and the improvement of communication technologies,massive amounts of data have been generated in all walks of life,how to dig out valuable information from mass moving object trajectory data becomes very important.Clustering analysis is the most commonly used and one of the most important technique in the process of data mining.It can be used to analyze and process trajectory information,so as to discover hidden relationships and rules in data,and predict the future movement of moving objects.It is the basis of further analysis and processing of data.The Classic trajectory,which describes the overall motion pattern of the trajectory in the cluster and can be regarded as a model of the trajectory cluster to some extent.Representative trajectory generation refers to that selects an existing trajectory from trajectory cluster or generates a new composite trajectory as a representative of the overall operation mode of the cluster.A large amount of useful information about trajectory cluster can be extracted by generating Representative trajectory from trajectory cluster and performing correlation analysis,which serves as the basis for trajectory prediction.To sum up,this paper hopes to design a Classic trajectory generation framework that is suitable for trajectory data in both restricted space and unrestricted space based on density peak clustering algorithm.The framework mainly includes three modules:1)Trajectory similarity distance matrix calculation module.The existing main trajectory similarity measurement methods are classified and their characteristics and applicable scenarios are analyzed firstly.Then According to the requirements of trajectory similarity distance measurement in this paper,considering the influence of trajectory noise points,an improved SSPDP distance is proposed based on SSPD distance to calculate the trajectory similarity distance matrix as the trajectory similarity distance measurement.2)Trajectory clustering module.As core modules of classical trajectory generation framework,it based on the analysis of the characteristics of the traditional clustering methods on the basis of combining trajectory shape is diversiform and have the features of a large number of noise points,so choose the density peak clustering as trajectory clustering algorithm.Then aiming at the existence drawbacks of the original density peak cluster,such as the local density calculation method cannot apply to different scale data sets and the data set to identify the sparse clusters,puts forward the modified clustering algorithm DPKNN-DPC based on density peak.3)Classic trajectory generation module.After the trajectory clustering is finished,the following is how to generate the classical trajectory from trajectory cluster,this paper puts forward three kinds of classical trajectory generation methods.The Best Representative Merge method does not generate a new optimal trajectory but pick up one from the existing track in clusters to other tracks of minimizing the sum of the overall similarity distance as the classical trajectory.The other two methods,Even Interval Merge and Sweep Line Approach generate a new synthetic trajectory as the classical trajectory.Finally,experiments are carried out on the trajectory data sets in the restricted space and unrestricted space.Then the quality of clustering is evaluated by BC index and WC index,which are compared with different classical trajectory generation frameworks to verify the accuracy and feasibility of the classical trajectory generation algorithm in this paper.