RGB-D Feature Selection, Model Training and Classifier Adaptation Strategies Evaluation for Object Category Recognition

Object category recognition is an important problem in many applications involving indoor robots. When equipped with latest sensors providing depth information besides traditional RGB colour images, such robots are in a better position to identify objects of interest. Several methods for object category recognition in RGB-D images have been reported in literature. These methods are typically tested under the same conditions (domain) such as viewing angles, distances to the object as well as lightening conditions on which they are trained. However, in practical applications one often has to deal with previously unseen domains. We explore the suitability of 3D and image-based features along with adaptation strategies to be able to successfully categorize novel objects during deployment, based on information learned from online databases.