Adaptation of cluster analysis methods to optimize a biomechanical motion model of humans in a nursing bed

Abstract

The paper considers the optimization of a Hidden-Markov Model (HMM) based method for the generation of averaged motion sequences. To create averaged motion sequences, motion sequences of different test persons were originally recorded with a motion capture system (MoCap system) and then averaged using an HMM approach. The resulting averaged data sets, however, partly showed serious motion artifacts and uncoordinated intermediate movements, especially in the extremities. The aim of this work was to combine only movements with similar courses in the extremities by a suitable cluster analysis. For each test person, model body descriptions of 21 body elements are available, each of which is represented in three-dimensional time series. For optimization, the MoCap data are first compared using time warp edit distance (TWED) and clustered using an agglomerative hierarchical procedure. Finally, the data of the resulting clusters are used to generate new averaged motion sequences using the HMM approach. The resulting averaged data can be used, for example, in a simulation in a multilevel biomechanical model.