Articulated hand pose estimation plays an important role in human-computerinteraction. Despite the recent progress, the accuracy of existing methods isstill not satisfactory, partially due to the difficulty of embeddedhigh-dimensional and non-linear regression problem. Different from the existingdiscriminative methods that regress for the hand pose with a single depthimage, we propose to first project the query depth image onto three orthogonalplanes and utilize these multi-view projections to regress for 2D heat-mapswhich estimate the joint positions on each plane. These multi-view heat-mapsare then fused to produce final 3D hand pose estimation with learned posepriors. Experiments show that the proposed method largely outperformsstate-of-the-art on a challenging dataset. Moreover, a cross-dataset experimentalso demonstrates the good generalization ability of the proposed method.
展开▼
