Clustered Pose and Nonlinear Appearance Models for Human Pose Estimation

• DOI: 10.5244/c.24.12
• Publication Date: 2010-08-31T07:08:29Z
• Authors: Sam Johnson; Mark Everingham
• Keywords: 0202 electrical engineering, electronic engineering, information engineering; 02 engineering and technology

We investigate the task of 2D articulated human pose estimation in unconstrained still images. This is extremely challenging because variation pose, anatomy, clothing, and imaging conditions. Current methods use simple models body part appearance plausible configurations due to limitations available training data constraints on computational expense. show that such severely limit accuracy. Building successful pictorial structure model (PSM) we propose richer both using state-of-the-art discriminative classifiers without introducing unacceptable introduce a new annotated database consumer images, an order magnitude larger than currently datasets, demonstrate over 50% relative improvement accuracy stateof-the-art method.