D.J. Jung, K.S. Kwon, and H.J. Kim (Korea)
Human pose estimation, Bayesian network, model parameter optimization
In this paper, we propose a method for human pose estimation from single camera images. The human body configuration is modeled by a Bayesian network. In this model, state nodes represent an aspect ratio, a length of long axis, a degree of tilt, a 2D center position and 2D position of body joints of each body parts and edges represent spatial constraint. We estimate the human pose through optimization of the model parameters based on simple person model. The model parameters are estimated from cues such as a color information, ratio of the length of long axes between node and parent node, and distance of body joints between node and parent node. For the estimation of dynamic human pose, we use temporal constraint. We apply the proposed method to the videos obtained from a gesture-based game control system and evaluate the performance using ground truth method. As a result, the proposed method can automatically label upper body pose from videos and it is appropriate method for feature extraction in the frontal view-based gesture recognition system.
Important Links:
Go Back
