Many dance poses involve the hand being in front of the body; these postures are particularly important in many mudras[+]. The binarized processing described above is not sufficient for resolving this occlusion, so the wrist is identified in the image based on the greyscale skin tones (see Figure 3). During the performance, for each grabbed image, the pixels outside this intensity range are discarded. If the user is wearing clothes contrasting with his body color, only the pixels corresponding to his body parts retain the high value of intensity. This leads to an effective separation of the hand from the body in cases of self-occlusion (see Figure 4). The hand can now easily be detected.
|
|
Figure 5: Image Processing Results. Left: Locating outermost tip. Right: Finding the elbow and wrist joints.
In the case where the hand is before the body, we still need to identify
the elbow for constructing the 3-D arm pose. This is simple here, since
the outermost tip of the image corresponds to the elbow joint.