Abstract
In this paper, we propose a method for estimating the pose of a human hand using normal histograms. The method is composed coast to fine of three steps. First, within the database which consist of hand model postures in a reduced degree of freedom space about human hand postures in a daily life, screening of postures is conducted using dispersion of point clouds. Second, the similarity between normal histograms generated from the input posture and the database posture is evaluated with Chi Square. Third, optimization for minimizing the evaluation function is conducted in the reduced parameter space. In the experiment, the hand model was used as input postures. The result of the experiment shows that the maximum estimation error was 21[%] using this method.
