Spatial Coverage Measurement of Geo- Tagged Visual Data: A Database Approach

Abstract

Due to the popularity of GPS-equipped cameras such as smartphones, geo-tagged image datasets are widely available and they became a good source to record every corner of urban streets and to understand people's everyday life. However, even in the presence of such large visual datasets, a simple question is not yet answered about how much such data cover a certain area spatially. For example, when we have millions of geo-tagged images in San Francisco, how do we know if this dataset visually covers the city completely or not in an intuitive way? Do we have visual coverage of a specific region from all directions or from only a certain direction? This paper provides an answer to such a question by introducing new measurement models to collectively quantify the spatial and directional coverage of a geo-tagged image dataset for a given geographical region. Our experimental results using large real datasets demonstrate that our proposed models are able to well represent the spatial coverage of a geo-tagged image dataset, which is comparable to human perception.