Tony Heap and David Hogg
School of Computer Studies,
University of Leeds,
Leeds, UK, LS2 9JT
ajh@comp.leeds.ac.uk
The Point Distribution Model (PDM) has proved useful for many tasks involving the location and tracking of deformable objects. A principal limitation is non-specificity; in constructing a model to include all valid object shapes, the inclusion of some invalid shapes is unavoidable due to the linear nature of the approach.
Bregler and Omohundro [2] describe a `piecewise linear' method for applying constraints within model shape space, whereby principal component analysis is used on training data clusters in shape space to generate lower dimensional overlapping subspaces. Object shapes are constrained to lie within the union of these subspaces, thus improving the specificity of the model.
This is an important development in itself, but its most useful quality is that it lends itself to automated training. Manual annotation of training examples has previously been necessary to ensure good specificity in PDMs, requiring expertise and time, and thus limiting the amount of training data that can feasibly be collected. The use of shape space constraints means that such accurate annotation is unnecessary, and automated training becomes significantly more successful.
In this paper we expand on Bregler and Omohundro's work, suggesting an alternative representation for the linear pieces, and showing how a two-level hierarchy in shape space can be used to improve efficiency and reduce noise. We perform an evaluation on both synthetic and (automatically trained) real models.