GeoVideolndex: Indexing for georeferenced videos

Abstract

Videos captured with spatiotemporal information such as time, location, and camera direction are called georeferenced videos. As recent video recording devices such as smart phones, action camcorders, and dashcams have built-in GPS sensors, they capture videos with spatiotemporal information, and such spatiotemporal information can be used for querying georeferenced videos. For a video search system supporting location queries, an efficient spatial index is important to find the query results fast while maintaining its size small. This paper proposes an efficient indexing method for searching georeferenced videos, called GeoVideolndex. GeoVideoIndex adopts MBTR(Minimum Bounding Tilted Rectangle) in leaf nodes, as an MBTR can efficiently represent the viewable areas of a camera along the trajectory. GeoVideolndex constructs MBTRs only based on the linear change of camera moving direction, in order to form a long MBTR covering a linear piece of the trajectory. In particular, GeoVideolndex applies a data compression technique, which excludes superfluous scenes and stores data in a compact form. We experimentally compared the performance of spatial indexing methods on both real and synthetic datasets, and GeoVideolndex substantially reduces the index size and the construction time. GeoVideolndex also processes location queries much faster than other methods as well as manages vast amount of scenes compactly. (C) 2016 Elsevier Inc. All rights reserved.