Semantic localization in the PCL library

Abstract

The semantic localization problem in robotics consists in determining the\nplace where a robot is located by means of semantic categories. The problem is\nusually addressed as a supervised classification process, where input data\ncorrespond to robot perceptions while classes to semantic categories, like\nkitchen or corridor.\n In this paper we propose a framework, implemented in the PCL library, which\nprovides a set of valuable tools to easily develop and evaluate semantic\nlocalization systems. The implementation includes the generation of 3D global\ndescriptors following a Bag-of-Words approach. This allows the generation of\ndimensionality-fixed descriptors from any type of keypoint detector and feature\nextractor combinations. The framework has been designed, structured and\nimplemented in order to be easily extended with different keypoint detectors,\nfeature extractors as well as classification models.\n The proposed framework has also been used to evaluate the performance of a\nset of already implemented descriptors, when used as input for a specific\nsemantic localization system. The results obtained are discussed paying special\nattention to the internal parameters of the BoW descriptor generation process.\nMoreover, we also review the combination of some keypoint detectors with\ndifferent 3D descriptor generation techniques.\n