A Realization of Mutual Information Calculation on GPU for Semi-Global Stereo Matching

Abstract

According to the Middlebury Stereo Database, Semi-Global Matching (SGM) is commonly regarded as the most efficient algorithm among the top-performing stereo algorithms. Recently, most effective real-time implementations of this algorithm are based on reconfigurable hardware (FPGA). However, with the development of General-Purpose computation on Graphics Processing Unit (GPGPU), an effective real-time implementation on general purpose PCs can be expected. As the major theoretical basis and the first step of Semi-Global matching algorithm, the efficiency of mutual information (MI) calculation is important for real-time application. In this paper, a realization of MI calculation on Graphics Processing Unit (GPU) is introduced. Some important optimizations according to Compute Unified Device Architecture (CUDA) are introduced in this paper.