Retrieving large-scale high density video target tracks from spatial database

With more and more live sensors being added to geospatial applications, huge amount of sensor data are generated and saved in spatial database. Managing and mining these large-scale ever-changing data becomes new challenges for geospatial studies. In this paper, we present an application-oriented case study to show how to retrieve target tracking data from big dataset saved in spatial database. Our video event retrieval system collects thirty days (8790 GB) high definition video data from six surveillance cameras, analyze them and extract roughly ten million video target tracks. These tracks are projected onto world coordinates and pumped into a spatial database. The system performance of inserting and retrieving these tracks is analyzed in terms of spatial data type design, spatial index configuration, online operation capacity, query optimization and scalability handling. Our insights of saving, managing and retrieving target tracks in a large-scale are presented.