Notes:
- original samples represented as B-splines
- stable tracking requires reduction of dimensions => PCA is
one way to do that
- shape space is a linear parameterisation => computationally
simple to use
Tracking based on Kalman filtering
Block diagram
Tracking example
Notes:
- position, alignment and shape parameters estimated separately
- linear filtering => computationally fast
- problem: measurement sensitive
Tracking based on multisampling filters
Principle (one-dimensional case)
Tracking example
Notes: - properly selected hypotheses matched to image data - resistant to clutter - problem: the number of samples may be large
