Hierarchical Disparity Estimation With Programmable 3D Hardware

This paper presents an implementation of a hierarchical disparity estimation algorithm entirely executed on programmable 3D graphics hardware. Since programmable 3D graphics hardware has increased its algorithmic and computational power, several research teams exploit the SIMD (single instruction, multiple data) facilities of 3D graphic processing unit (GPU) to accelerate applications not related to rendering and visualization. Using graphics hardware for image processing and computer vision tasks is very natural, since graphics hardware is optimized to access and manipulate 2D images.