A Cloud-Based Trajectory Index Scheme

The “dead space” effect produced by the typical minimum bounding rectangle (MBR) based range search decreases the spatial discrimination when the number of trajectories increases and will be getting worse when the average coverage of trajectories enlarged. Consequently, a more straightforward index scheme is proposed to increase the spatial discrimination and keep the response time reasonable at the same time. This paper elaborates on the design of the trajectory index scheme and it is experimented with practical data gathered from a Web service, StreetImage 2.0, in which users can upload videos taken while wandering on the streets, with or without GPS logs. Those videos will be clipped into a sequence of images and associated with streets on road network. A sequence of images is called a ‘trail’ in this service and it can be shared with other users via keyword search, range search or a direct link. Currently, thousands of trails, implying millions of coordinates, extend along streets with higher density in downtown and lower in countryside. To perform a range search is a burden in server side. With the help of “Cloud”, large amount of jobs can be distributed over a cluster of processors with the illusion of infinite computing resource available on demand. We build our index scheme on Hadoop and HBase, and deploy the static index to the ‘Cloud’ for handling considerable requests of range search. The experiments show that an average response time to complete a range search is around 1.7 seconds and it took an average of 52 milliseconds to get the first response.