The SB-ST decomposition in the study of Developmental Coordination Disorder

To deal with redundancy and high-dimensionality that are typical of movement data, we propose to decompose action matrices in two decoupled steps: first, we discover a set of key postures, that is, vectors corresponding to key relationships among degrees of freedom (like angles between body parts) which we call spatial basis (SB) and second, we impose a parametric model to the spatio-temporal (ST) profiles of each SB vector. These two steps constitute the SB-ST decomposition of an action: SB vectors represent the key postures, their ST profiles represent trajectories of these postures and ST parameters express how these postures are being controlled and coordinated. SB-ST shares elements in common with computational models of motor synergies and biological motion perception, and it relates to human manifold models that are popular in machine learning. We showcase the method by applying SB vectors and ST parameters to study vertical jumps of adults, typically developing children and children with Developmental Coordination Disorder obtained with motion capture. Using that data, we also evaluate SB-ST alone and against other techniques in terms of reconstruction ability and number of dimensions used.