This study proposes a direction-based similarity measure for trajectory clustering. The proposed description of the trajectory was based on extracting the direction changes in the segmented trajectories (sub-trajectories). The authors applied spectral clustering to segment a trajectory to several sub-trajectories. Then, trajectory descriptions were computed based on the direction change in different levels of resolution in terms of trajectory instances. To measure the similarity of trajectories, these segments were used as the input of Time Warp Matching method. Finally, the hierarchical clustering was applied to cluster similar trajectories. The direction-based description helps to achieve rotation and location invariance characteristics. Some experiments were performed to compare the proposed trajectory descriptor with similar approaches in the application of trajectory clustering. The empirical quality of the proposed similarity measure is evaluated on a clustering task. Compared to well-known similarity measures, the proposed method proved to be effective in the considered experiment.
