Alignment of Similar Shapes Based on their Convex Hulls for 3D Object Classification

In order to facilitate true autonomous robot manipulation, object classification becomes inevitable part of advanced robotic vision. Since objects belonging to the same class usually have similar shape, we propose a new method for object alignment based on similarity of their convex hulls. In this paper, the purpose of the developed alignment method is to generate hypotheses for object classification, and return pose of the classified object. The proposed approach consists of generating alignment proposals and evaluation of these proposals by a method commonly used in shape instance detection algorithms. The proposed approach is verified on an object classification benchmark dataset, where the proposed method achieves competitive results in comparison to four state-of-the-art object classification methods.