Tracking with Context as a Semi-supervised Learning and Labeling Problem

Supplementary Materials

This page contains supplementary materials to the paper Cerman, L., Hlaváč, V.: Tracking with Context as a Semi-supervised Learning and Labeling Problem accepted to ICPR 2012.

All presented videos have two parts. The first part demonstrates tracking window initialization and the second part demonstrates the tracking/labeling results. The videos are played in slower (15 FPS) than real (25 FPS) frame rate to better see what is happening.

For better video quality, please switch the embedded YouTube player to HD resolution (720p).

Phone sequence

This is the video presented in the submitted paper.

The video can be also downloaded as a video file here (26,2 MB).

Phone sequence (70 frames window)

This is the same video but with a tracking window extended to 70 frames. It shows that extending the windows allowed to successfully track the object until the end of the sequence.

The video can be also downloaded as a video file here (25,9 MB).

Cup

In the proposed algorithm, a planar motion (2-D homography) is assumed. This would suggest only planar objects can be tracked. However, modeling the residual noise provides sufficient robustness to the deviations from a homography to track such non-planar objects as cups are.

The video can be also downloaded as a video file here (19,1 MB).

Hands

This video shows the importance of the motion modeling. Using only appearance features the tracked hand would not be distinguishable from the other hands.

The video can be also downloaded as a video file here (20,8 MB).

Glove

This additional video shows tracking of a glove.

The video can be also downloaded as a video file here (17,3 MB).

Box

This additional video shows tracking of a pill box.

The video can be also downloaded as a video file here (20,3 MB).

Last Updated on 8-Apr-2012