A direction based framework for trajectory data analysis

We propose a framework for the directional analysis of trajectory data. The directional aspect of trajectory analysis is important in map matching, in direction based query processing and in animal movement data. The main contribution in the present work lies in the trajectory segmentation method which is based on directional changes in trajectory. Another contribution is the use of convex hulls of trajectories during filtration of outlier sub-trajectories. There are four components to the framework: (1): smoothing, (2): directional segmentation and classification, (3): outlier subtrajectory filtering and (4): clustering. We split the trajectories into directional sub-trajectories such that they have a specific directional characteristics; for example, heading north-east. We consider 16 directional classes and assign the corresponding directional sub-trajectories to them. In the filtration step the outlier sub-trajectories are removed from the respective directional classes using a novel convex hull based approach. We compare convex hull filtering performance with conventional minimum bounding rectangle based approach. We finally cluster the filtered directional sub-trajectories to obtain global directional patterns in the data set using a modified DBSCAN algorithm. We also provide the comparison of proposed work with an existing state-of-the-art algorithm called TRACLUS. In this work two real data sets are analyzed: hurricane data and animal movement data.