Research On Trajectory Clustering Algorithms Of Moving Objects

With the continuous development of computer technology as well as the continuous improvement of moving object tracking technology, massive trajectory data has been collected. In order to discover the hidden knowledge of these data, the moving object trajectory clustering technique is produced as the times require. As an important research branch of data mining, cluster analysis divide the objects into some close but independent clusters according to the the principle that objects with maximum similarity are in the same cluster and objects with minimum similarity are in the different cluster. In this paper,we take the sub-trajectory clustering algorithm of moving object as the research direction and aim at the defects of TRACLUS algorithm, we have conducted some research and exploration by considering the algorithm from the clustering factor and improving the sensitivity of clustering algorithm to parameters. The main work is as follows:Aiming at the TRACLUS algorithm has not considered the problem of the motion characteristics of the trajectory when the similarity measure is carried out, we propose a new trajectory clustering algorithm HDBSCAN based on fusion loss and hausdorff distance by taking the direction feature of the moving object into consideration. By applying the algorithm to the real trajectory data, we know that the algorithm has a better clustering effect while ensuring the efficiency of TRACLUS algorithm.Aiming at the problem that TRACLUS algorithm is sensitive to the parameters ?and MinLns, we propose a fast search algorithm HFDST for high density sub-trajectory clustering. The algorithm uses the partition and grouping framework. We also combine the fusion loss and hausdorff distance together to measure the distance between the sub-trajectory, and a method of fast searching clustering center published in science is applied to the sub-trajectory clustering. The clustering center is defined as the local maximum density point and the density is only determined by the distance between the sub-trajectory. This algorithm can not only detect the Non sphere clustering, it can also automatically find the number of clusters. So HFDST can overcome the dependence of TRACLUS algorithm on parameters. By applying the algorithm to the real trajectory data, we know that the algorithm has a very good clustering effect,the time efficiency of the algorithm is greatly improved and it has good robustness.