Fast visual trajectory analysis using Spatial Bayesian Networks

During the past years the first tools for visual analysis of trajectory data appeared. Considering the growing sizes of trajectory collections, one important task is to ensure user interactivity during data analysis. In this paper we present a fast, model-based visualization approach for the analysis of location dependencies in large trajectory collections. Existing approaches are not suitable for visual dependency analysis as the size and complexity of trajectory data constrain ad hoc and advance computations. Also recent developments in the area of trajectory data warehouses cannot be applied because the spatial correlations are lost during trajectory aggregation. Our approach builds a compact model which represents the dependency structures of the data. The visualisation toolkit then interacts only with the model and is thus independent of the size of the underlying trajectory database. More precisely, we build a Bayesian Network model using the Scalable Sparse Bayesian Network Learning (SSBNL) algorithm, which we improve to represent also negative correlations. We implement our approach into the GIS MapInfo using MapBasic scripts for the user interface and an independent mediator script to retrieve patterns from the model. We demonstrate our approach using mobile phone data of the city of Milan, Italy.