Computational Considerations on Randomized Hough Transform and Motion Analysis

A new and eecient version of the Hough Transform, the Randomized Hough Transform (RHT), has recently been suggested. The RHT has been applied to motion detection by the author and his coworkers and a novel method to calculate 2-D motion in a sequence of time-varying images has been proposed. The method, called Motion Detection using Randomized Hough Transform (MDRHT), utilizes the main ideas of the RHT, random sampling of, e.g., edge points of the original image, and a converging mapping of randomly sampled points into one point in a dynamically-linked accumulator. In this paper the generalized algorithm is suggested and computational behavior of this generalization is examined. Probability mechanisms of the MDRHT are considered on both translation and rotation.