OpenCL Implementation of a Color Based Object Tracking
University of Novi Sad, Faculty of Technical Sciences, Novi Sad, Serbia
3rd International Conference on Information Society Technology (ICIST), 2013
@article{jocic2013opencl,
title={OpenCL Implementation of a Color Based Object Tracking},
author={Joci{‘c}, Marko and Obradovi{‘c}, {DJ}or{dj}e and Konjovi{‘c}, Zora and Tertei, Daniel},
year={2013}
}
In this paper we present an algorithm for realtime object tracking based on color. Firstly, a two-layer perceptron is trained aimed at coping with scene illumination changes. Based on this training, a piece of OpenCL code is generated for the purpose of harnessing the power of GPU computing. Then, color based object tracking is done in four steps, and finally a fast connected component labeling algorithm is applied for determining distinct regions, with only the largest one selected and tracked on the image. Proposed algorithm is executed both on CPU and on GPU for comparative analysis.
June 12, 2013 by hgpu
