Starting from our experience in the field of 3D objects recognition from videos, in this project we study and develop methods for robotic applications.
We have at disposal a stereo vision system able to observe an object of interest for some time and changing the viewpoint.
We are currently studying method of data-driven dictionary learning to be applied to the problem of object recognition from videos frame per frame. We consider training videos in which only one object of interest is present and model their content to infer objects models. Such models are then used to perform the recognition at run-time, when more than one (possibly unknown) object may be observed.
This research is carried out with the Italian Institute of Technology (IIT) of Genova.
- Delponte, E. et al. "The importance of continuous views for real-time 3D object recognition". ICCV07 Workshop on 3D Representation for Recognition (3dRR-07), 2007.
- Noceti, N., E. Delponte and F. Odone. "Spatio-temporal constraints for on-line 3D object recognition in videos". Computer Vision and Image Understanding 113 (2009): 198-1209.