String-Based Feature Representation for Trajectory Clustering

Literature Review

Trajectory data is regarded as time sequenced spatial data and a number of trajectory clustering methods have been proposed and a review of these methods is discussed by Zheng (2015) and Sabarish, Karthi and Gireeshkumar (2015). Trajectory clustering approaches analyse trajectories as a whole as a single unit or as multiple segments. Gaffney and Smyth (1999) proposed a model-based clustering approach where trajectories are considered to be generated from mixture models and EM algorithm is used to assign data to clusters. Trajectories are considered as a whole and assigned to clusters based on cluster parameters. Lee, Han, Li and Gonzalez (2008) proposed traclus algorithm that generates many line segments for a trajectory using characteristic points in partition phase and clusters line segments using dbscan algorithm in grouping phase. Trajectories are considered as multiple sub trajectories and helps in discovering patterns in these sub trajectories.

Tripathi, Debnath and Elmasri (2016) proposed a framework where directional orientation of the trajectories is used for clustering. The algorithm does smoothing, filtration, segmentation and clustering of segments to obtain overall directional patterns in the data set. Mao, Zhong, Qi, Ping and Li (2017) proposed an adaptive dbscan trajectory clustering method where trajectory are segmented into sub trajectories using modified Minimum Description Length. The clustering space is divided into grids and trajectories are mapped into grids and an adaptive dbscan algorithm is proposed for clustering. A survey of trajectory clustering algorithms is discussed by Yuan, Sun, Zhao, Li and Wang (2017) where the trajectory algorithms are classified into categories as spatial based, time based, group and partition, uncertainty, sematic based, road network and optimization based methods for clustering.