Probabilistic Tracking in a Metric Space
Probabilistic Tracking in a Metric Space
Kentaro Toyama
and
Andrew Blake
This paper was presented at
ICCV 2001.
Final versions of the paper:
Below are videos to accompany the paper:
-
Pedestrian, generated sequence, using dynamics only:
generatd.mpg.
-
Pedestrian tracking, train and test on same person (different sequences):
walk1.mpg.
-
Pedestrian tracking, train on one person, test on another:
walk3.mpg.
-
Pedestrian tracking, train on one person, test on another, with simulated occlusion:
walk3occ.mpg.
-
Mouth tracking, train and test on same person (different sequences),
metric is L-2 image distance:
ml2.mpg.
-
Mouth tracking, train and test on same person (different sequences),
metric is shuffle distance:
mshuffle.mpg.
Preliminary results on larger exemplar sets:
-
Tracking a ballet dancer (training and test sequence are the same):
ballet0.mpg.