Hypothesis Generation in Generic, Model-Based Object Recognition System
One of the key problems in robotics is ability to search and recognize objects, as well as their reliable localization. Existing object-detection solutions relies either on instance-models, making them unable to generalize, or can’t estimate object pose if the model is not known at hand. We aim at making system able to work with generic models, described using set of simple parts and relations between them, as well as localize them on scene. Article presents hypothesis generation part of the designed system.