Learning stick-figure models using nonparametric Bayesian priors over trees

We present a probabilistic stick-figure model that uses a
nonparametric Bayesian distribution over trees for its structure
prior. Sticks are represented by nodes in a tree in such
a way that their parameter distributions are probabilistically
centered around their parent node. This prior enables
the inference procedures to learn multiple explanations for
motion-capture data, each of which could be trees of different
depth and path lengths. Thus, the algorithm can
automatically determine a reasonable distribution over the
number of sticks in a given dataset and their hierarchical
relationships. We provide experimental results on several
motion-capture datasets, demonstrating the model’s ability
to recover plausible stick-figure structure, and also the
model’s robust behavior when faced with occlusion.