Synthesizing Transition Paths among Multiple Motion Models for Human Pose Tracking

Norimichi UKITA  Noriaki ICHIDA  Masatsugu KIDODE 

Publication
D - Abstracts of IEICE TRANSACTIONS on Information and Systems (Japanese Edition)  Vol.J94-D  No.8  pp.1260-1268
Publication Date: 2011/08/01
Online ISSN: 1881-0225
Print ISSN: 1880-4535
Type of Manuscript: Special Section PAPER (Special Section on Image and Recognition and Understanding)
Category: 
Keyword: 
human pose estimation,  dynamical model of the human pose,  dynamical models of multiple actions,  switching the dynamical models,  transition paths between the dynamical models, 

Full Text(in Japanese): PDF(4.3MB)

Summary: 
This paper proposes a method for estimating the pose of a human body by employing motion prior of multiple actions. Motion prior is obtained independently for each action such as walking and jogging. This independent modeling allows us 1) to improve the scalability of the method and 2) to optimize each model in accordance with its motion. The models, each of which is used for predicting an input motion at each moment, are automatically switched when the motion of a subject changes between the actions. The delay of this switch occurs loss of followability in pose tracking. The proposed method finds prospective transition points among different models based on similarity of the models, and then the transition paths between the prospective points are virtually generated. These paths enable reactive transitions between different actions. Experimental results demonstrate improvement of the reactivity and its effectiveness for accurate pose tracking.