Share Email Print

Proceedings Paper

Online recognition of group actions in intelligent meeting scenario
Author(s): Xiang Zhang; Linmi Tao; Guangyou Xu
Format Member Price Non-Member Price
PDF $14.40 $18.00

Paper Abstract

Group actions play a key role in intelligent meetings. In this paper, we propose a probabilistic approach to online segment meetings as a sequence of group actions, such as monologue, presentation, discussion, and break. In our approach, we decompose group actions into three sub-actions according to three sorts of features independently: audio features, video features and group visual features. In accordance with this assumption state spaces are decomposed into two levels of resolution: meeting actions and meeting sub-actions. Multi-stream dynamic Bayesian network is constructed based on three level state nodes modeling group actions, sub-actions, and three sorts of features. Particle filters are applied to efficiently online recognize group actions, which is based on the estimate of joint posteriors over node states of multi-stream dynamic Bayesian network. Posterior probabilities over all state spaces are represented by temporal sets of their weighted samples. We make seven compared experiments with the different sample numbers of 50,100, 200,300,400, 500 and 600. The recognition accuracy gets higher when there are more sample numbers, but is takes more time for event inference.

Paper Details

Date Published: 28 November 2007
PDF: 8 pages
Proc. SPIE 6833, Electronic Imaging and Multimedia Technology V, 683312 (28 November 2007); doi: 10.1117/12.760170
Show Author Affiliations
Xiang Zhang, Yangtze Univ. (China)
Tsinghua Univ. (China)
Linmi Tao, Tsinghua Univ. (China)
Guangyou Xu, Tsinghua Univ. (China)

Published in SPIE Proceedings Vol. 6833:
Electronic Imaging and Multimedia Technology V
Liwei Zhou; Chung-Sheng Li; Minerva M. Yeung, Editor(s)

© SPIE. Terms of Use
Back to Top