FPGA Implementation of the Generalized Hough Transform

Many applications of image analysis need a similarity measure that highlights the presence of shapes or objects. The Generalized Hough Transform (GHT) is a popular image processing technique that is able to locate an object in an image. In its original formulation, this algorithm has a regular computation pattern, but suffers from high computational and memory requirements. More efficient GHT implementations have been carried out leading to computation and memory saving at the expenses of introducing irregularities in the computation, which make more difficult the design of a specific hardware solution. This work proposes the use of Field-Programmable Gate Arrays (FPGAs) for the implementation of an efficient version of the GHT. The development of the GHT has been divided into several functional blocks. This permits us to take advantage of a progressive reduction of the data flow and the algorithm stages, in order to optimize the use of the FPGA resources and clock cycles. We have tested our design by applying the GHT to the similarity calculation of frames in a video sequence.