PIPELINED IMPLEMENTATION OF THE MULTIRESOLUTION HOUGH TRANSFORM IN A PYRAMID MULTIPROCESSOR

The Hough transform is used to detect patterns in images. Its advantage is the ability to detect discontinuous patterns in noisy images. The drawback is its demand for a large amount of computing power. Efficient algorithms and their implementations in multiprocessor systems have been proposed to overcome the computational requirements. The ability to efficiently may two-dimensional (2D) images into 2D meshes makes the pyramid computer a very attractive candidate for implementation of image processing algorithms. Several implementations of the Hough transform in pyramid machines can be found in the literature. The implementations are geared towards the analysis of a single image. Moreover, the implementations do not consider the balancing of the computational load uniformly among the processor elements in the pyramid. The objective of this paper is to propose an efficient pipelined implementation of the multiresolution Hough transform in pyramid machine. The proposed implementation considers task scheduling to ensure balanced computational load among the processor elements in the pyramid. It is shown that the proposed pipelined implementation is suitable for efficiency analyzing a continuous sequence of images, as found in a typical industrial inspection system. The multiresolution images and accumulator arrays used in the multiresolution Hough transform (MHT) mirrors the variable sized 2D processor arrays in the pyramid layers. The MHT has therefore, been used for implementation in the pyramid machine.