Human Body Tracking with Shape Estimation and Auxiliary Measurements

A real time 3D human body reconstruction is performed using two to four synchronized cameras. Silhouettes of the detected regions are extracted and registered allowing a 3D reconstruction of the human body using Generalized Cylinders. When four cameras are used simultaneously, we derive from the detected silhouettes the visual-hull of the human body shape. A global shape descriptor is computed from the polygonal representation of the visual-hull allowing the recognition of 13 human body postures across a variety of people. We use a machine learning approach (support vector machine) for training the system to recognize various body postures from the perceived visual-hull.

Video [quicktime] | [winmedia]

3D Body Modeling Technology Summary
3D Body Postures Technology Summary
NSF Report
Poster
More Info
Laboratory