Faster Upper Body Pose Estimation and Recognition Using CUDA
Department of Computer Science, University of the Western Cape, Private Bag X17 Bellville, 7535, South Africa
Southern African Telecommunication Networks and Applications Conference (SATNAC), 2012
@article{brown2012faster,
title={Faster Upper Body Pose Estimation and Recognition Using CUDA},
author={Brown, Dane L and Ghaziasgar, Mehrdad and Connan, James},
year={2012}
}
Image processing techniques can be very time consuming when applied linearly on the Central Processing Unit (CPU). Many applications require processing to take place in real-time. The Upper Body Pose Estimation and Recognition system developed by Achmed and Connan has shown to be 88% accurate, but operates at less than real-time on the CPU. This paper proposes an adapted version of this algorithm, which runs on the Graphics Processing Unit (GPU) to achieve real-time processing speed. The system was found to achieve a slightly improved recognition accuracy of 92.95% while achieving on average a real-time processing speed of no less than 18 Frames Per Second (FPS) and a mean speed of 33.26 FPS.
April 25, 2013 by hgpu
