Research On Trajectory Similarity Measure Based On Movement Features

With the rapid advances of positioning (e.g. BeiDou navigation satellite system, GPS) and wireless communication technologies, people have collected a huge amount of spatio-temporal trajectory data of moving objects. Spatio-temporal trajectory data mining has become a research hotspot and plays a significant role in intelligent transportation system, climate monitoring, movement ecology and other fields. Movement features are the characteristics of moving parameters like speed and acceleration, they can reflect the inherent characteristics of moving objects and the external environment’s impact on the behaviors of moving objects. This thesis studies trajectory similarity measures based on movement features, this is a core problem in trajectory data mining and can be used in similarity search and movement pattern discovery.This thesis utilized moving objects’ movement features as the main line and studies the trajectory similarity measures in-depth. Through improving and developing novel trajectory similarity measures based on movement features, the study applies the similarity measures to related applications. The main research work and results are summarized as follows:1, On the basis of the summary of existing movement parameters and movement features, the study proposes the similarity measure based on hierarchical movement features and classification learning methods and applies the measure to moving objects recognition. The similarity measure extracts global and local movement features from trajectories to form hierarchical movement features. Global feature are extracted with advanced statistics and local features are extracted with movement parameter time series segmentation. Hierarchical movement features are integrated with the classification method—supporting vector machine (SVM) to construct the measure. The experiments on three real trajectory datasets show that the discriminative power of the measure is large and it improves the moving objects recognition accuracy significantly compared with existing methods.2, The study proposes multiple movement features edit distance (MMFED) as the trajectory similarity measure based on multiple movement features and applies the measure to movement sequential pattern discovery based on multiple movement features. MMFED borrows the idea of data cube, quantizes and symbolizes the multiple movement parameter time series. In multiple movement feature domain space, the Euclidean distances between characters are computed as the substitution costs of weighted edit distance, and the weighted edit distance is computerd as the similarity measure. The measure reflects the evolution trend of multiple movement features, i.e. movement sequential pattern. The measure is integrated with the spectral clustering method for movement sequential pattern discovery. Using the Atlantic hurricane dataset, the known hurricane originating and movement laws in meteorological literatures verify the effectiveness of the measure. Moreover, the hurricanes’ movement sequential pattern is analyzed.3, The study proposes the trajectory spatio-temporal similarity measure fused of movement features and applies the measure to the discovery of spatio-temporal distribution pattern based on movement features. The similarity measure fuses spatial distance, temporal distance and movement feature distance. The measure employs EDR (Edit Distance on Real Sequence) as the spatial distance, NWED (Normalized Weighted Edit Distance) as the movement feature distance and uses the start points’distance, the end points’ distance and the durations of the trajectories to measure the temporal distance. Finally, through weighted average, the measure fuses the three distances together. The measure is integrated with the spectral clustering method for the discovery of spatio-temporal distribution pattern based on movement features. Using hurricane dataset, the known hurricanes’ spatial and seasonal laws in meteorological literatures verify the effectiveness of the similarity. Moreover, the spatio-temporal distribution law of hurricanes’ speed is analyzed.The thesis improves and develops novel trajectory similarity measures based on movement features. Furthermore, the study boosts the theoretic and applied research on trajectory similarity measures and provides valuable research results for trajectory data mining.