Quantitative Movement Analysis Based on a Hierarchical Spatial-temporal Correspondence of Movement Primitives

We present a new learning-based approach for the representation of complex motion patterns that is based on the linear combination of prototypical movement sequences that can be used in clinical gait and movement analysis. Opposed to many existing methods that are based on heuristically chosen individual kinematic or kinetic features, this method extracts automatically the meaningful parts of the spatio-temporal characteristics of whole movement trajectories and maps the trajectories into an abstract topological space with axes that have a simple meaningful interpretation. We give an overview over the algorithm and describe examples for the analysis, quantification and synthesis of movements. Further, we describe clinical applications on the quantitative analysis of patients suffering from Parkinson disease.