Anatomical Joint Constraint Modelling with Rigid Map Neural Networks

Abstract

The development of anatomical models both for individuals and groups are important for applications in animation, medicine and ergonomics. Recent approaches have utilised unit quaternions to represent orientations between limbs which eliminate singularities encountered in other rotational representations. As a result a number of unit quaternion based joint constraint validation and correction methods have been developed. Recent approaches harness machine learning techniques to model valid orientation spaces and has included the use of Kohonen's Self Organizing Maps (SOMs) to model regular conical constraints on the orientation of the limb. Recent work has considered a derivative of the SOM, the Rigid Map, applied in the same context which we extend here.