Research On Trajectory Clustering Algorithms Of Moving Objects

With the development of computer technology as well as continuous improvement of moving object tracking technology, massive trajectory data is collected. It becomes urgent to extract hidden information from these data through effective analysis, which leads to the generation and development of moving object trajectory clustering technology. Trajectory clustering of moving object can find objects with the same or similar movement behavior, and extract their common feature pattern, using which to analyze and predict the future moving behavior of moving objects.For the purpose of finding common sub-trajectory of moving objects, in this paper, research and exploration are carried on in views of the weakness of TRACLUS. The scope of this research lies in some burning issues, which involve in enhancing TRACLUS’s executing efficiency and shielding its parameter sensitivity and so on. The main contributions of this paper are summarized as follows:Firstly, existing spatial index structures can not be directly used to support efficient neighborhood query of line segment in TRACLUS’s line segment clustering phase, making the time complexity of line clustering algorithm is O(N2), where N is the total number of line segments after trajectory partitioned, it requires large volume of memory support and needs a lot of I/O costs as N increases. Aim to this issue, reference line is introduced to trajectory clustering, and RLTC based on the reference line is presented, which uses a certain number of reference lines to represent the spatial characteristics of a line clustering approximately. The clustering of the line segments can be implemented by the classification of reference lines. The experimental results demonstrate that RLTC maintains the trajectory clustering result of TRACLUS while improves the efficiency.Secondly, aiming to the problem that TRACLUS is sensitive to the input parameters such likeεand MinLns, a trajectory clustering algorithm named OPSTC is proposed, which can shield clustering algorithm’s parameter sensitivity. OPSTC defines the core distance and reachability distance of line segment, according to which generates processing sequence of line segment, which actually contains the information that equals to the line segment clusterings based on the density from a broad inputting range of parameters. Then it constructs the rechability plot of line sets according to the sequence, with which extracts the final clustering structures after proper distance is selected for extraction. The experimental results reveal that OPSTC reduces TRACLUS’s dependency and sensitivity to input parameters effectively, meanwhile ensures the quality of trajectory clustering.