Combination of Classifiers for Indoor Room Recognition CGS participation at ImageCLEF2010 Robot Vision Task

摘要

Abstract. This paper represents a description of our approach to the problem of topological localization of a mobile robot using visual information. Our method has been developed for ImageCLEF 2010 Robot Vision Task challenge. The challenge was focused on the problem of visual place classification, with a special focus on generalization. The goal was to recognize rooms by the images captured with a stereo camera mounted on a mobile robot within an office environment. Algorithms should be able to reply to question “Where are you?”, saying “I don’t know ” if the room analyzed was not presented during training phase. For the challenge three sequences were given: Training Set, Validation Set and Test Set acquired on three different floors of the same building. We chose to approach the challenge realizing a multi-Level machine learning architecture, made of a first “weak ” classifiers Level based on visual features extracted from images and of a second Level performing fusion of first Level outputs. We developed four configurations to determine the best approach to