Simplify shapeDescriptor dependencies

[gsaponaro]

Currently, shapeDescriptor requires these image inputs:

• camera colour image: this is only needed for hue colour histogram computation (do we really need those? -> Tony did) and for overlay image output (it looks nice but it should be optional -> done);
• binary segmented image: necessary;
• labelled image (0=background, 1=object, 2=another object, 3=another object, etc.): it should be computed internally by the module, doing the sequential labelling in a thread, rather than with an external module. We could even make labelled images redundant altogether, by re-thinking the data structures and feature extraction.

Also, it could be useful to exploit the ROI information (list containing the blobs bounding boxes, corresponding to the boundingRectangle descriptor) to further reduce the code.