Tracking the human body (figs. from [TomRaider, Fanti08, Gall09]).

MSc dissertation proposal 2010/2011

Skeleton Tracking

Introduction:

The game market recently introduced cameras able to reconstruct the pose of a person or, more precisely, the pose of a simplified skeleton model. The triggering element allowing this technological feat was the introduction of an high tech laser range finder component having the shape of a matrix of sensors inside a conventional imaging sensor.

Supervision: Prof. José Gaspar and Prof. Alexandre Bernardino.

Objectives:

The objectives of this work are twofold: (i) studying the methodologies for skeleton tracking based on video and laser range finding, (ii) identifying applications tacking advantage of observing the full kinematic chain of human bodies.

Detailed description:

Estimating the 3D motion of humans or animals is a fundamental problem in many applications, including realistic character animation for games and movies, or motion analysis for medical diagnostics and sport science [Gall09]. This long standing research and industrial objective, used to require specific electromagnetic emitters / detectors, many cameras and/or specific clothing [Polhemus], is getting now into the gaming market, as a natural (marker-less) interface, with very competitive prices, such as the Microsoft Kinect camera [MS-Kinect].

Fig. The Kinect camera, Microsoft
(fig. from [MS-Kinect]).

The base hardware in the Microsoft Kinect camera comprises a laser range finder (actually a matrix of sensors) and video cameras. The superimposed video on the range images allows than to distinguish whether the silhouette of a hand is closer or further away relative to the body. By trying to discriminate the various locations of the body members, the device ends up reconstructing the kinematic chain composing a human body.

In this project the main objective is to study the problem of human body skeleton tracking, using video and range finder cameras.

References:

[TombRaider] http://www.tombraider.com

[Fanti08] C. Fanti. "Towards Automatic Discovery of Human Movemes", PhD thesis, Caltech, 2008.

[Gall09] "Motion Capture Using Joint Skeleton Tracking and Surface Estimation", Juergen Gall, Carsten Stoll, Edilson de Aguiar, Christian Theobalt, Bodo Rosenhahn, and Hans-Peter Seidel, in CVPR 2009.

[Polhemus] http://www.polhemus.com/

[MS-Kinect] http://en.wikipedia.org/wiki/Kinect

Expected results:

At the end of the work the students will have enriched their knowledge in:

* Computer vision

* Tracking of articulated objects

Observations:

--

More MSc dissertation proposals on Computer and Robot Vision in:

http://omni.isr.ist.utl.pt/~jag