Tracking with Adaptive Appearance Models
The WSL-appearance model is described in the CVPR 2001 paper:
A.D. Jepson, D.J. Fleet, and T.F. El-Maraghi,
Robust online appearance models for visual tracking,
IEEE CVPR, 2001.
The Dudek face sequence (html)
is now available on-line. That page includes a link to the 1145 frame
sequence used in the CVPR '01 paper above, along with seven ground
truth points located by hand for benchmarking purposes.
This ground truth data was first used in:
A.D. Jepson, D.J. Fleet, and T.F. El-Maraghi,
Robust online appearance models for visual tracking,
submitted IEEE PAMI, 2002.
The following MPEG files show tracking results using the
WSL-appearance model:
-
The Dudek face sequence shows tracking in the presence of
partial occlusion and also
appearance changes due to changes in the 3D pose, changes in
facial expression.
MPEG (9MB)
-
The Fleet equip sequence shows tracking in the presence of
significant and persistent partial occlusion.
MPEG (2MB)
-
The Fleet face sequence shows tracking in the presence of appearance
change. Near the end of the sequence the subject stops and turns.
The appearance model then learns the background, and when the subject
starts to move again the tracking fails.
MPEG (5MB)
-
The Fleet car sequence shows tracking in the presence of appearance
change as the subject rotates 180 degrees, resulting in a large variation
of appearance. The end of the sequence also demonstrates persistent
total occlusion, which causes the tracking to fail.
MPEG (6MB)