A Real-World System for Image/Video Geo-localization

Abstract

Determining where an image was taken and geo-locating depicted structures are important tasks from a surveillance and intelligence standpoint. For example, the image might show terrorist training facilities or the vicinity of a safe house. To geo-localize, the user must combine prior knowledge of the area with subtle clues from the image in order to mitigate the tedious manual search of GIS reference data. This process is extremely challenging, time-consuming, and often yields poor accuracy. In this chapter, we describe WALDO (Wide Area Localization of Depicted Objects), a system that solves this challenging problem by combining the insight of analysts with the power of automated analysis for Internet-scale, geo-location-driven data mining. WALDO's goal-driven constrained resource management leverages a full spectrum of data-driven, semantic, and geometric geo-localization experts and user tools.