2010 Poster Sessions : Real-time Motion Capture using a Single Time-Of-Flight Camera

Student Name : Varun Ganapathi
Advisor : Daphne Koller
Research Areas: Artificial Intelligence
Abstract:
Markerless tracking of human pose is a hard yet relevant problem. In this paper, we derive an efficient filtering algorithm for tracking human pose at 10 frames per second using a stream of monocular depth images. The key idea is to combine an accurate generative model---which is achievable in this setting using special-purpose yet standard graphics hardware---with a discriminative model that feeds data-driven evidence about body part locations to the filter. In each filter iteration, we apply a form of local model-based search that exploits the nature of the kinematic chain. As fast movements and occlusion can disrupt the local search, we utilize a set of discriminatively trained patch classifiers to detect body parts. We describe a novel algorithm for propagating this noisy evidence about body part locations up the kinematic chain. The resulting distribution of body configurations allows us to reinitialize the model-based search, which in turn allows our system to robustly recover from temporary tracking drift.

Bio:
Varun is a PhD student at Stanford University studying machine learning and artificial intelligence. His primary research focus is the use of probabilistic models in computer vision. Varun is co-advised by Daphne Koller and Sebastian Thrun. Varun received an MS in CS and BS in physics from Stanford in 2006 and 2005. He also enjoys computational photography and, with a colleague, created an application for the iPhone, "Pro HDR" that can combine two differently exposed images to produce one high dynamic range image.