Towards Robust and Accurate Similar Trajectory Discovery: Weak-Parametric Approaches

Abstract

Trajectory analysis is crucial and has been more and more widely used in various fields, such as location-based services (LBS), urban traffic control, user classification and route planner, etc. In this paper, we propose GSIM and ASIM, two novel approaches that are weak-parametric and can effectively measure and discover similar trajectories. The proposed methods are based on the key insight that the similarity can be reflected by observing the growth rate of specific indicators. (1) GSIM defines a 3-layer grid structure and statistics the total overlapping points for all grids between trajectories in each layer, it finally calculates the growth rate of the total counts as the grid radius grows from layer 1 to layer 3. (2) ASIM assumes that any two trajectories are similar and calculates the area of the minimum boundary rectangle that contains all the points. Then it cuts the rectangle from four directions one point by one to get the maximum boundary rectangle that contains the other two percentage of total points. Finally it utilizes the average change rate of the areas as the similarity. Further, we design parameter-learning modules to learn the setting of corresponding parameters automatically. Extensive experiments on real-world dataset show that, compared with typical approaches like LCSS, EDIT, DTW, etc., the proposed methods can significantly improve the effectiveness and achieve better efficiency in most test cases. Meanwhile, they are not sensitive to parameter settings.