OROnto: An Ontology for Recognition of Grasping Objects

Abstract

This paper reports development of an ontology using Web Ontology Language (OWL), focused to the task of object recognition by prosthetic hands. The ontology named as OROnto (Object Recognition Ontology) is comprised of attributes, concepts and relationships among the human hand grasp types and structural entities of objects. After validation by a reasoner, cases have been presented in this work where the inferred ontology was able to retrieve object types against the user’s Description Logic (DL) queries. A grasp experiment was performed to study the effectiveness of the ontological attributes towards object recognition. Classification results using data from semantically suggested features showed a 2-4% higher recognition accuracy in comparison to the results using data with the features selected by the popular random forest based feature selection method. This reveals that apart from the extraction of implicit and explicit information of the domain knowledge, ontologies can also be used as a feature selection method for classification problems.