USING FOCUS OF ATTENTION WITH THE HOUGH TRANSFORM FOR ACCURATE LINE PARAMETER-ESTIMATION

In this paper we describe a Hough transform algorithm for finding lines in an image that can have an order of magnitude denser sampling in both parameters of the parameter space without significant increase in required memory or overhead in CPU time. By increasing the sampling density we determine the line parameters for each line segment much more accurately. The method circumvents the problems of requiring large amounts of memory to store the accumulator array, and is not significantly slower than other Hough algorithms by using a focus of attention approach. We illustrate the benefits that can be obtained on a wide variety of images, and quantify the improvements in line parameter estimation by performing statistical tests. We also discuss the benefits that can be obtained by algorithms attempting to locate higher level features in the image, and determining the 3D model of the image scene.