Simultaneous 3D human shape estimation and motion tracking is a very challenging problem. Given the complicated nature of human shape and motion, it can be very difficult to perform robust tracking of body parts without using a motion or shape model. Modelling the human body as rigid parts linked in a kinematic structure is a simple yet reasonably accurate model for tracking purposes. Optical flow can be exploited to provide dense information and obtain robust estimates of the motion parameters. Bregler and Malik (1998) use orthographic projection, while Sidenbladh et al. (2000) use a Bayesian formulation combined with a particle filtering approach to determine the motion parameters. Yamamoto et al. (1998) use a different set of motion parameters to perform tracking. They use multiple views and perspective projection in their model but have a large parameter set and make approximations in their formulation.
[1]
David J. Fleet,et al.
Stochastic Tracking of 3D Human Figures Using 2D Image Motion
,
2000,
ECCV.
[2]
Takuya Kondo,et al.
Incremental tracking of human actions from multiple views
,
1998,
Proceedings. 1998 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (Cat. No.98CB36231).
[3]
Jitendra Malik,et al.
Tracking people with twists and exponential maps
,
1998,
Proceedings. 1998 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (Cat. No.98CB36231).
[4]
Richard M. Murray,et al.
A Mathematical Introduction to Robotic Manipulation
,
1994
.