Some methods for minimising computation with the Hough transform

This paper studies the amount of computation involved in Hough transform approaches to object and feature location. In general, it is found that most of the computation involves building parameter space rather than searching it for peaks. A simple formula predicts a worst case for a certain degree of feature complexity, and all Hough-based procedures must avoid this in one way or another. Different approaches have different characteristics with regard to image clutter and accuracy of object location, as well as speed of computation. Attention is focussed on the sensitivity and accuracy of object location resulting after a limited number of complex features have been employed