Selected Readings in Vision and Graphics
edited by Luc Van Gool, Gábor Székely, Markus Gross, Bernt Schiele

Volume 51

Tobias Jaeggli

Statistical Models for Human Body Pose Estimation from Videos

2008. XVI, 151 pages. EUR 64,00. ISBN-10: 3-86628-235-2

ISBN-13: 978-3-86628-235-3

Recently, statistical Machine Learning methods have become popular in Computer Vision. In this thesis, their application to the visual analysis of human body motion is investigated. The presented algorithms aim at finding and following humans in video sequences, and at computing a 3D estimate of the person's body posture and motion, i.e. the configuration and movements of the rigid body parts. This specific task has many practical applications, and has been chosen as an example task for the general problem setting of estimating multi-dimensional continuous vectors from video data.
Statistical models of prior knowledge about human motion, and of the relation between body poses and the visual appearance of humans are learned from a set of training data. Given previously unseen video sequences, body poses are then estimated by analytical and sample-based probabilistic inference strategies, resulting in posterior probability distributions that reflect remaining uncertainties and ambiguities of the estimation task.
The algorithms were tested on challenging real sequences with low-resolution and noisy imagery. Finally, extensions are presented, that allow for dealing with multiple persons that are simultaneously visible in the images, and for recognising the category of the performed activity.

About the Author:
Tobias Jaeggli studied computer science with a minor in media and communication sciences at the University of Fribourg (CH), where he graduated with a MSc in 2002. He then worked as a research assistant in computer vision at the Katholieke Universiteit Leuven (B), and subsequently started his PhD studies at the ETH Zurich in 2004. In 2008, he was awarded a PhD degree (Dr. sc. ETH Zürich), for his work about statistical models for visual human pose estimation.

Keywords / Schlagwörter:
Human Pose Estimation, Statistical Models, Machine Learning, Monocular Tracking, Computer Vision

Reihe " Selected Readings in Vision and Graphics " im Hartung-Gorre Verlag

Direkt bestellen bei / to order directly from: Hartung.Gorre@t-online.de