Analysis of the efficiency of the census transform algorithm implemented on FPGA

Over the course of the last two decades, continuous advances in the stereo vision field have been documented. In this paper we present an analysis of the efficiency for the stereo vision algorithm of the Census Transform algorithm. In addition to the conventional correlation method based on Hamming distance minimization, we use two similarity measures: the Tanimoto and the Dixon-Koehler distances. Then, we compare its performance in terms of accuracy and hardware resources needed for implementation. These comparisons are performed by introducing a generalized model for each hardware architecture, scalable depending on design parameters such as Census Transform window size and maximum disparity range.