Recognition of 3-D objects with a closest point matching algorithm

This paper is a contribution to the recognition of arbitrary 3-D shapes from range images. Recently, a closest point matching algorithm was proposed that is in a position to match 3-D shapes without the need of an explicit object description in terms of primitives. The advantage is that the comparison works directly on the 3-D coordinates of the surface points and needs neither segmentation nor feature extraction. Arbitrary shapes can potentially be matched with it. In this paper we present a number of 3-D object recognition experiments conducted for assessing the capability of this approach to recognise objects measured by range images. The experiments performed with the designed methods show the fast convergence and reliable recognition of objects. We also address some of the key issues of this matching algorithm, which are the strong dependence on the initial conditions, like position of the object to be recognised, and the quadratic computing complexity.