Interpolated SOM Neural Networks for Anatomical Joint Constraint Modelling

Unit quaternions offer a singularity free representation when modelling orientations between limbs at a joint. The development of accurate joint constraint models for such joints is a non-trivial task and several approaches have been suggested including a number which leverage machine learning which aim to create joint models based training data from an individual or group. Previous work has demonstrated the use of an extended Rigid Map Neural Network with a continuous output to model conical constraints (with a regular boundary). In this paper we employ a similar approach deploying an extended Self Organising Map (SOM) with a continuous output.