Hand tracking with an extended self-organizing map

Abstract

We introduce an extension of the self-organizing map for performing 3D hand skeleton tracking. We use a range camera for data acquisition and apply a SOM-like learning process within each frame in order to capture the hand pose. Our method uses a topology consisting of 1D and 2D segments for an improved representation of the hand. The proposed algorithm is very efficient and produces good tracking results. © 2013 Springer-Verlag.