A novel algorithm for automatic 3D model-based free-form object recognition

Recognition of free-form objects from unknown viewpoints is a difficult task, especially in the presence of occlusions. In this paper we address this problem and present a novel algorithm for automatic 3D model-based free-form object recognition. We use a robust tensor-based representation for matching surface patches from the scene with a model library. A 4D hash table is built from the tensors during the offline phase in order to make the online matching efficient. Our algorithm can accurately identify objects in occluded environments and calculate their pose while matching the scene and the model at a very low resolution. Preliminary experiments on a library of real objects show that our algorithm is applicable to free-form objects, accurate, efficient and robust to occlusions.