A sampling approach to ultra-fast object location

This paper examines how sampling can form the basis for ultra-fast location of objects in digital images. The body-based sampling approach that is ado pted is unusual in that detection speed is improved for larger objects. The fact of sampling leads initially to more imprecise object location; this p oses the problem of finding methods by which object location can be refined without prejudicing the speed of the initial rapid search. In the case of circular objects this problem is straightforward to solve: it is more taxin g in the case of ellipses. but a highly effective procedure has been found in the "triple bisection" algorithm, which is simply applied by the bisecti on of chords across the elliptical objects. These techniques have been appl ied successfully to the location of cereal grains, which can be modelled as ellipses with a 2:1 aspect ratio. However, the main purpose of this paper is to show that ultra-fast object location is possible with off-camera imag es, and to indicate the ultimate speeds that are attainable by these means. (C) 2001 Academic Press.