The abstraction for trajectories with different numbers of sampling points

Trajectory abstraction is an efficient way to handle the large amount of information included in complex trajectory data. Based on the previous work, this paper proposes an improved framework for abstracting trajectories, which consists of three major stages. First, the original trajectories in different lengths are matched into groups according to their similarities, and then a non-local denoising approach, based on the wavelet thresholding technique, is performed on these groups to summarize trajectories. Last, a combined version of the compacted trajectories is obtained as the final trajectory abstraction. To avoid loss of trajectory features introduced by the resampling technique, we provide a novel method to convert trajectories in different lengths into suppositional equal, which serves for the similarity measurement and the wavelet thresholding. Extensive experiments on real and synthetic trajectory datasets demonstrate that the proposed trajectory abstraction achieves very potential results dealing with complex trajectory data.