Starting from our experience in the field of 3D objects recognition from videos, in this project we study and develop methods for providing the iCub, a humanoid robot designed at IIT, with visual perception skills. In particular we address the problem of understanding the scene semantics, starting from the ability of learning to recognize objects or object categories.
As a distinctive part of our research we developed a Human-Robot-Interaction (HRI) schema where the iCub can label images automatically exploiting self-supervision by either using knowledge on its own kinematics or visual motion cues. This knowledge provides a prior on the object location in the image which facilitates data gathering and further processing. This platform allowed us to gather the iCubWorld datasets, which are available for download More information on the iCubWorld can be found here. An example of the data acquisition procedure can be seen at: http://www.youtube.com/watch?v=vhPLUNg9r5k
Based on the availability of the data and of an inspiring platform such as the iCub, we studied the problem of learning sparse representations for effective image classification and object recognition. Starting from a consolidated visual recognition pipeline we propose novel ideas on coding and pooling procedures. For a quick taste of our research have a look at this video http://www.youtube.com/watch?v=ZIpVrD6e-kA
This research is carried out with the Italian Institute of Technology (IIT) of Genova and the Laboratory for Computational and Statistical Learning (IIT@MIT).
- Fanello, Sean Ryan et al. "iCub World: friendly robots help building good vision data-sets". Computer Vision and Pattern Recognition Workshops. IEEE, 2013.
- Fanello, Sean Ryan et al. "Multi-class image classification: sparsity does it better". VISAPP: International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications.. Barcelona, Spain, 2013.
- ---. "Dictionary based pooling for object categorization". VISAPP: : International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications., 2014.