ACCELERATION OF KEY FRAME EXTRACTION AND TRACKING ALGORITHMS THROUGH GENERAL PURPOSE GPU COMPUTING USING OPEN CL PROGRAMMING FRAMEWORK

Abstract

Modern processor architectures have embraced parallelism as an important pathway to increased performance. Now, Central Processing Units (CPUs) improve performance resulted by adding multiple cores. Graphics Processing Units (GPUs) have also evolved from fixed function rendering devices into programmable parallel processors. As today’s computer systems often include highly parallel CPUs, GPUs and other types of processors, to take full advantage of these heterogeneous processing platforms, OpenCL (Open Computing Language) provides the new way of computing. OpenCL plays a significant role in emerging interactive graphics applications which integrates general parallel computing algorithms with graphics rendering pipelines. Here GPU computing is applied on General Purpose applications that are Key Frame Extraction and Tracking Algorithms with the help of OpenCL. In order to retrieve a particular piece of information in a video, of late, Video summarization, aimed at reducing the amount of data that must be examined and that also becomes an essential task in applications of video analysis and indexing. Generally, a video summary is a sequence of still or moving images, with or without audio. Our work is mainly based on acceleration of one such algorithm that utilizes visual summary using still images, called key frames, extracted from the video. Here advantages of still images is that it can summarize the video content in more rapid and compact way, so users can grasp the overall content more quickly from key frames than by watching a set of video sequences. In our case we optimized the pre-processing algorithms for image refinement using Frequency Selective Weighted Median Filter (FSWM) and feature extraction using histogram calculation to accelerate the Key Frame Extraction (KFE) algorithm. The optimization is done through general purpose GPU computing using OpenCL programming framework. Other part of our work is related to the acceleration of the feature tracking algorithms. As the ability to reliably detect and track human motion is a useful tool for higher-level applications, such as image analyzer that rely on visual input, interacting with human activities are at the core of many problems in intelligent systems, such as humancomputer interaction and robotics. Our work focuses on how to speed up the process of KLT (Kanade-Lucas-Tomasi) tracking and how to utilize advantages of FAST (Features from Accelerated Segmented Test) algorithm in the KLT tracking. The algorithm (FAST and KLT) selects the features that are optimal for tracking and keeps the track of these features.