Spatio-temporal constraints for matching view-based descriptions of 3D objects