A Parallel Framework for Silhouette-Based Human Motion Capture

This paper presents a method to capture human motion from silhouettes of a person in multi-view video streams. Applying a hierarchical kinematic body model motion parameters are estimated by optimizing the overlap between the projected model silhouettes and the input silhouettes. The energy function driving the optimization is computed very efficiently using off-the-shelf graphics hardware. Exploiting the hierarchical structure of the human body, energy function evaluation is greatly sped up and a distributed implementation becomes feasible. Therefore, we present an algorithm for parallel silhouette-based motion capture employing multiple PCs and GPUs.