Feature points selection for markerless hand pose estimation

One of the conditions for accurate planar pose estimation is that feature points must be both coplanar and noncollinear. Many research on markerless hand tracking and pose estimation as a planar target have been done, however the selection of hand feature points as coplanar but noncollinear points has not been investigated. This paper proposes a novel selection of hand feature points for pose estimation that improves the pose estimation. Markerless hand pose estimation as a continuous tracking of rigid planar object is made possible using robust planar pose (RPP) algorithm implemented on a marker-based Augmented Reality Toolkit (ARToolkit) library. The results obtained show significant improvement over recent approaches on the accuracy of the estimated pose such as in the rotation and the translation parameters and pose ambiguity problems are greatly reduced.