A New Hough Transform for the Detection of Arbitrary 3-Dimensional Objects
John McDonald, David S. G. Vernon · Maynooth University ePrints and eTheses Archive (Maynooth University) · 1998
. The existing Generalised Hough Transform, although altered to cater for scaling and rotation of the object in the plane, fails to detect the object under rotations out of the plane. This is due to the lack of 3dimensional information contained in the 2-dimensional template image. In this paper we present a new Hough Transform, known as the Surface Normal Hough Transform (SNHT), which using a suitable 2-dimensional surface representation, transforms a set of surface normals to a surface parameter space. The effect of the SNHT is to map point sets representing a surface in the input space, to a peak in the parameter space. The coordinates of this peak parameterise the given surface and hence allow for pose invariant object detection and localisation. Keywords: 3-D Computer Vision; Hough Transform; Pose Invariant Object Detection; Surface Registration. 1 Overview It is generally accepted that the main aim of computer vision is to realise an adaptive system which is capable...