Publications

Abstract : Articulated body pose estimation in computer vision is an important problem because of convolution of the models. It is useful in real time applications such as surveillance camera, computer games, human computer interaction etc. Feature extraction is the main part in pose estimation which helps for a successful classification. In this paper, we propose a system for extracting the features from the relational graph of articulated upper body poses of basic Bharatanatyam steps, each performed by different persons of different experiences and size. Our method has the ability to extract features from an attributed relational graph from challenging images with background clutters, clothing diversity, illumination etc. The system starts with skeletonization process which determines the human pose and increases the smoothness using B-Spline approach. Attributed relational graph is generated and the geometrical features are extracted for the correct discrimination between shapes that can be useful for classification and annotation of dance poses. We evaluate our approach experimentally on 2D images of basic Bharatanatyam poses.