Result

In this section, various issues arising from the implementation of CONDENSATION based visual tracking system are discussed and some results from the implementation are presented. For the experiment, a simple video clip is taken by Intel PC Pack Camera. Since the spatial/temporal resolution is very low(160 * 120, 15 frames/sec), the scene was carefully taken so that the observation module can correctly detect the image features in the image sequence. The result shows that the system is very successful in tracking the contour of the object, which is a table tennis racket.

• Initialization
• Random Sampling from the Effective Prior - 
• Deterministic Drift from the Dynamic Model - 
• Observation Model and Approximate Posterior - 

 
• Here are the Matlab files used in the implementation and the resulting tracking sequence.


Matlab codes

• condensation.m
• computeWeightFromObservation.m
• doCurveFitting.m
• driftByDynamicModel.m
• findNormals.m
• generateStatesFromInitial.m
• observeImageFeatures.m
• randomSamplingFromPrior.m


Matlab Data

• trackingData.mat


Tracking Video File

• trackingDemo.AVI

Tue Mar 14 02:05:32 PST 2000