Home // UBICOMM 2014, The Eighth International Conference on Mobile Ubiquitous Computing, Systems, Services and Technologies // View article

Action Recogntion with Depth Maps Using HOG Descriptors of Multi-view Motion Appearance and History

Authors:
DoHyung Kim
Woo-han Yun
Ho-Sub Yoon
Jaehong Kim

Keywords: Action recognition; Depth maps; Depth motion appearance; Depth motion history; Histogram of oriented gradients.

Abstract:
The goal of this work is to recognize human actions only using depth maps without additional joints information. As a practical solution, we present a novel volumetric representation of global shape of depth motion, Depth Motion Appearance (DMA). The proposed framework also extracts dynamic information of the body movements called Depth Motion History (DMH), an extended version of motion history image. In the framework, a huge amount of data of an action video is summarized into concise action representation maps observed from multi-view. A histogram of oriented gradients then describes local appearances and shapes of the DMAs and DMHs, which results in more compact and discriminative action representation. The presented method has been compared with the state-of-the-art approaches on a public dataset. The experimental result demonstrates that our approach achieves a better and more stable performance with a relatively smaller feature maps and lower complexity.

Pages: 126 to 130

Copyright: Copyright (c) IARIA, 2014

Publication date: August 24, 2014

Published in: conference

ISSN: 2308-4278

ISBN: 978-1-61208-353-7

Location: Rome, Italy

Dates: from August 24, 2014 to August 28, 2014