Multi-source Spatio-temporal Similarity Algorithm Based On Symbolic Representation Of Trajectory Movement Parameters

With the development of sensor technology to record the multi-dimensional motion state of moving objects,a large amount of motion behavior trajectory data has been generated.By measuring the similarity of these trajectory data,it is of great significance to study the movement behavior patterns between different moving objects.Although a large number of trajectory similarity algorithms have emerged,they generally suffer from the following shortcomings:Most of them are restricted to geometric abstractions of the objects' movement path as a static curve,and only a few of the available similarity analysis techniques take the variations of movement parameters into account.However,In multi-dimensional and complex research scenarios,spatial similarity alone is not enough to determine movement.These methods become unsuitable when there are similar motion features between objects.Moreover,In movement behavior studies,it is essential to take into account the key parameters that characterize the movement of objects,such as speed,acceleration,or direction.However,In those works that apply the trajectory movement parameters to the similarity calculation,they either directly use the movement parameter values to abstract the trajectory,or only use some local features of the movement parameters to label the trajectory characters for research,hence they don't have considering the complex local features and global characteristics contained in the variations of movement parameters.And they all only use the movement parameters as a source of information for research,which will lose the complex spatiotemporal characteristics contained in the original trajectory.To solve the problems as stated,thesis proposes a multi-source spatio-temporal edit distance algorithm that fuses the changes of trajectory movement parameters and the temporal and spatial geometric characteristics of the trajectory.More specifically,MSTED mainly has two modules: trajectory segmentation and similarity algorithm.In the trajectory segmentation module,the trajectory is divided into uniform segments containing complex movement patterns according to the change of movement parameter attributes over time,and eighteen categories are divided according to the movement parameter attributes to convert the original trajectory into a sequence of character labels.In the similarity algorithm module,thesis expands the traditional edit distance,specifically,which integrates the movement pattern information contained in the trajectory movement parameters and the original spatiotemporal geometric characteristics of the trajectory to design the cost functions of the three operations of edit distance.Finally,thesis performs extensive experiments on two public real datasets,and the experimental results show the effectiveness of each component in MSTED and the overall MSTED.