BENCHMARKING OF MULTIMEDIA APPLICATIONS ON OPEN CL PLATFORM

Abstract

Khronos Group created the OpenCL standard that allows using full compute power of a computer. The goal of this thesis is to show how to utilize the current advantages of OpenCL language by implementing motion estimation algorithm. Motion estimation is the process of determining motion vectors that describe the transformation from one 2D image to another usually from adjacent frames in a video sequence. In this project we have implemented & tested different algorithms for achieving faster performance for generating motion vectors between two consecutive frames .This is a part of MCTI technology for generation of best quality video. Motion vector generation has been done on forward & backward matching of frames and is being tested on C & OpenCL on different algorithms. We have achieved very interesting and surprising results by varying sequential code to parallel code on GPUs. We have achieved 16ms per frame and have tested our code on two GPU cards 1. Nvidia Quadro NVS 295 2. Nvidia GTX 285 Thanks to an accessible programming environment and nearly full support of the C programming language, a large number of established APIs for images and pixel operations and bilinear prediction has already been ported to run on GPUs.