SIFT feature extraction on a Smartphone GPU using OpenGL ES2.0

Abstract

SIFT describes local features in image used for object recognition in a vast array of application, such as augmented reality, panorama stitching. These applications are becoming very popular on Smartphones but also require considerable amount of computing power. GPUs offer a significant amount of untapped computing power that can help increase performance and improve user experience. We explore the feasibility of parallel heterogeneous computing on current generation of Smartphone. We show that the CPU and GPU can work in tandem to solve complex problems. However the mobile platform remains very restrictive requires a lot of effort from the programmer but does not achieve the same performance gains as observed on the PC.