Deep Trajectory Similarity Model: A Fast Method for Trajectory Similarity Computation

Measuring trajectory similarity is a fundamental problem in the trajectory data mining field, and many similarity measurement methods had been proposed, such as dynamic time wrapping (DTW). However, these methods are dynamic programming problems, and dynamic programming problem usually leads to quadratic computational complexity. Thus, many acceleration algorithms were proposed. In this article, the authors proposed a deep neural network (DNN) based supervised similarity model, deep trajectory similarity model, to fit DTW similarity and to keep accuracy and orderliness. In the training process, they used low-frequency GPS trajectory data in Beijing as input data and used the DTW similarity of trajectory pairs as labels. In the test process, the model predicted the DTW similarity between two GPS trajectories. Experiments in this article indicated that deep trajectory similarity model could greatly decrease over 20% computation time than the acceleration algorithm of DTW similarity, FastDTW algorithm, and keep over 90% accuracy and over 97% orderliness. Experiments result indicated that the DTSM model has great potential in big data scenario.


  • English

Media Info

  • Media Type: Web
  • Features: References;
  • Pagination: pp 13-23
  • Monograph Title: International Conference on Transportation and Development 2019: Innovation and Sustainability in Smart Mobility and Smart Cities

Subject/Index Terms

Filing Info

  • Accession Number: 01732565
  • Record Type: Publication
  • ISBN: 9780784482582
  • Files: TRIS, ASCE
  • Created Date: Aug 28 2019 3:01PM