Abstract
With the increasing number of mobile GPS devices, more and more trajectory data are collected, stored, and analyzed for various applications. One of the basic operations in trajectory analysis is the similarity query, which retrieves similar trajectories of a given one. In this thesis, we model trajectory data as vector fields, by which the similarity between two trajectories can be measured in the vector space that they are transformed to. We call this algorithm of similarity measure for trajectory data Cosine Similarity for Vector Filed (CSVF). With such model, trajectory queries can be performed efficiently using Locality Sensitive Hashing (LSH) to filter out most dissimilar trajectories. Experiments which use Geolife dataset demonstrate that LSH can filter out nearly 70% candidate trajectories while maintaining the recall close to 98%. Meanwhile, experiments show that CSVF is 100 times faster than the tradition Longest Common Subsequence(LCSS) method when querying ten thousands of trajectory data.