Understanding and Visualisation of Geographic Mesh Similarity by Trajectory Data and Gaussian Process Modelling

A new concept is proposed of estimating mesh similarity based on trajectory data. The model is formulated as an unsupervised learning method using a type of Gaussian process on a continuous coordinate system. This allows for the features of meshes and trajectories to be determined as the estimated latent coordinates that are different from geographic ones. The similarities of meshes and trajectories are represented through those of coordinates. In addition, this allows for easy visualisation. After introducing the coordinate estimation method with a type of Markov Chain Monte Carlo approach, the proposed method was verified using actual trajectory data from the city of Sendai, Japan.