Human pose estimation with implicit shape models

We address the problem of articulated 2D human pose estimation in natural images. A well-known person detector -- the Implicit Shape Model (ISM) approach introduced by Leibe et al. -- is shown not only to be well suited to detect persons, but can also be exploited to derive a person's pose. Therefore, we extend the original voting approach of ISM and let all visual words that contribute to a person hypothesis also vote for the positions of the person's body parts. Since this approach is not constrained to a certain feature type and different feature types can even be fused during the pose estimation process, the approach is highly flexible. We show preliminary evaluation results of our approach using on the public available HumanEva dataset which comprises ground-truth pose data and thereby provides training and evaluation data.