A discriminative approach to improvements of indoor robot localization

Abstract

Indoor robot localization is a challenging problem in scene recognition. Generally, appropriate image representation and multiclass classifier are the two keys to the success of such a task. In this paper, a discriminative approach is proposed to meeting the challenges, which is composed of two steps: (1) spatial pyramid match and a Pyramid of HOG (Histograms of Oriented Gradient) are incorporated to represent an indoor place image. (2) a multi-stage SVM (Support Vector Machine) is utilized to classify an image by a cascade of one-versus-all SVMs. The proposed method achieves high accuracy on the ImageCLEF2012 and ImageCLEF2013 Robot Vision database, which shows the effectiveness of the proposed method.