In the last decade competitions proved to be a very efficient way of encouraging researchers to advance the state of the art in different research fields in artificial intelligence. In this paper we focus on the optional task of the RobotVision@ImageCLEF competition, which consists of a visual place classification problem where images are not isolated pictures but a sequence of frames captured by a camera mounted on a mobile robot. This fact leads us to deal with this problem not as stand-alone classification problem, but as a problem of self localization in which the robot’s main sensor only captures visual information. Thus, we base our proposal on a clever combination of Monte-Carlo-based self-localization methods with optimized versions of scale-invariant feature transformation algorithms for image representation and matching. The goodness of our approach has been validated by being the winners of this task in the 2009 RobotVision@ImageCLEF and 2010 RobotVision ImageCLEF@ICPR competitions.
Computer vision; Robot localization; Place recognition; Semantic place representation