Rendering articulated figures from examples

This paper presents new methods to robustly track and interpolate images of complex articulated figures. Traditional interpolation approaches fail on these cases since appearance is not necessarily a smooth function nor a linear manifold. We show how to model such appearance mappings with sets of examples which are each smooth functions. These well-behaved groups can cover overlapping regions of parameter space, and are empirically validated. A set growing procedure is used to find example clusters which are well-approximated within their convex hull; interpolation then proceeds only within a single such set. With this method physically valid images are produced even in regions of parameter space where the overall mapping is singular and nearby examples have different appearances. Our analysis stage tracks the occluding boundaries of articulated figures, even when there is no foreground contrast on an individual limb. Results generating both simulated and real arm images are demonstrated