We propose a method for tracking articulated body. In our framework, the
articulated tracking is considered as a problem of simultaneously tracking
multi-objects, where for instance, each object can be an arm, a leg, or a body
etc. in human motion tracking. However, due to the nature of articulated
structure, those moving parts usually move with less degree of freedoms
than regular independent objects. Borrowing the idea of particle filtering,
we sample candidates with weights in our problem of multi-object tracking.
E.g., a candidate can be a potential possibility of a waving arm at a moment
in a location. Different from the regular particle filtering, the weight is also
affected by how much the samples obey the motion constraints associated
with the articulated structure. The whole framework is divided into several
steps. The preprocessing step is aimed at finding those articulated parts. We
randomly sample moving pixels and use Gaussian mixtures to group them
into parts. The results from optical flow and locations are considered as
the input for the Gaussian mixtures. It is close to the abstract concept of
object tracking where a moving object is roughly defined as the set of points
with proximity and with the “common fate”. After the articulated parts are
obtained, we track them simultaneously by a particle filtering approach, with
weights adjusted according to the motion constraints. The whole process can
be done close to real time.

We propose a method for tracking articulated body. In our framework, the
articulated tracking is considered as a problem of simultaneously tracking
multi-objects, where for instance, each object can be an arm, a leg, or a body
etc. in human motion tracking. However, due to the nature of articulated
structure, those moving parts usually move with less degree of freedoms
than regular independent objects. Borrowing the idea of particle filtering,
we sample candidates with weights in our problem of multi-object tracking.
E.g., a candidate can be a potential possibility of a waving arm at a moment
in a location. Different from the regular particle filtering, the weight is also
affected by how much the samples obey the motion constraints associated
with the articulated structure. The whole framework is divided into several
steps. The preprocessing step is aimed at finding those articulated parts. We
randomly sample moving pixels and use Gaussian mixtures to group them
into parts. The results from optical flow and locations are considered as
the input for the Gaussian mixtures. It is close to the abstract concept of
object tracking where a moving object is roughly defined as the set of points
with proximity and with the “common fate”. After the articulated parts are
obtained, we track them simultaneously by a particle filtering approach, with
weights adjusted according to the motion constraints. The whole process can
be done close to real time.

[1] J. L. Barron, D. J. Fleet, and S. S. Beauchemin. Performance of optical
flow techniques. International Journal of Computer Vision, 12(1):43–77,
1994.
[2] M. Figueiredo and A.K.Jain. Unsupervised learning of finite mixture
models. IEEE Transaction on Pattern Analysis and Machine Intelligence,
24:381–396, 2002.
[3] M. A. Fischler and R. C. Bolles. Random sample consensus: A paradigm
for model fitting with applications to image analysis and automated
cartography. Comm. of the ACM, 24:381–395, 1981.
[4] I. Haritaoglu, D. Harwood, and L. Davis. W4 : Who? when? where?
what? a real time system for detecting and tracking people. In Proc.
IEEE Int’l Conf. on Face and Gesture Recognition, 1998.
[5] Berthold K. P. Horn and Brian G. Schunck. Determining optical flow.
Artificial Intelligence, 17(1-3):185–203, 1981.
[6] Michael Isard and Andrew Blake. Contour tracking by stochastic propagation
of conditional density. In Proceedings of the Fourth European
Conference on Computer Vision, 2:343–356, 1996.
[7] Michael Isard and Andrew Blake. Condensation conditional density
propagation for visual tracking. International Journal of Computer Vision,
29(1):5–28, 1998.
[8] K. Nummiaro, E. Koller-Meier, and L. Van Gool. An adaptive colorbased
particle filter. Image and Vision, 21(1):99–110, 2003.
[9] V. Pavlovic, J. M. Rehg, T. J. cham, and K. P. Murthy. A dynamic
bayesian network approach to figure tracking using learned dynamic models.
IEEE International Conference on Computer Vision, 1:94–101,
1999.
[10] H. Sidenbladh, M.J. Black, and L. Sigal. Implicit probabilistic models
of human motion for synthesis and tracking. ECCV, 1:784–800, 2002.
[11] Carlo Tomasi. Estimating gaussian mixture densities with em - a tutorial.
2003.
[12] C. Wren, A. Azarbayejani, T. Darrel, and A. Pentland. Pfinder: Realtime
tracking of the human body. IEEE Trans. on Pattern Analysis and
Machine Intelligence, 9:780–785, 1997.
[13] Ying Wu, Gang Hua, and Ting Yu. Tracking articulated body by dynamic
markov network. Proceddings of the Ninth IEEE International
Conference on Computer Vision, 2003.