3D object recognition by indexing structural invariants from multiple views

Abstract

We present a method for 3D object recognition from 2D image sequences. The system uses feature points tracked over three or more views to compute structural invariants, which serve as 3D shape representations. Object recognition is performed by using these Euclidean invariants as indices into a high-dimensional shape table. The use of indexing obliviates any need for matching models to images. In addition, the representation of 3D objects is extracted from 2D views, eliminating the cumbersome burden of having to obtain 3D models. The proposed scheme was implemented using a mixed database of real and simulated objects. We present experiments that show good recognition results on real objects and simulated objects corrupted with noise.