# Analysis of 3D motion patterns of male and female top-level handball players with Kohonen’s Self-Organizing Maps

Serrien, Ben and Blondeel , Jonathan and Clijsen, Ron and Baeyens, Jean Pierre (2013) Analysis of 3D motion patterns of male and female top-level handball players with Kohonen’s Self-Organizing Maps. In: 18e VK symposium "Bewegen met vallen en opstaan" (Moving with falling and getting up again), Friday 13.12.2013, FaBer-KU Leuven, Tervuursevest 101, 3001 Leuven, Begium.

 Text (Analysis of 3D motion patterns of male and female top-level handball players with Kohonen’s Self-Organizing Maps.) ppt_VKsymposium_dec2013-2.pdf Download (155Kb)

## Abstract

Introduction: Self-organizing maps as described by Kohonen (1) can be. used as a means for data reduction of high-dimensional data sets with possible underlying relations between the variables. Especially in human motion analysis they are useful because of their non-linear properties and will therefore outperform an analysis with classical PCA. Therefore we used this algorithm for the analysis of 3D data of a standing throw by handball players. We looked for differences between male and female players on a more holistic level in their throwing pattern. Traditional statistical analysis with data from discrete points in time is limited both in the linear approach as in the chance for a type 1 error with many variables. Materials and Methods: Three-dimensional coordinates of 11 male and 10 female Swiss handball players were collected with a VICON motion system (7 camera’s). These were transformed to Euler/Cardan angles in Mathcad according to ISB guidelines and angular velocities were calculated. This gave us 25 time series of 50 data points which were linearly normalized to a mean of 0 and a variance of 1 as the SOM uses Euclidean distances and the raw data are differently scaled. A SOM was made based on 1 reference trial of a random male player in the SOM-Toolbox for Matlab (2), and a U-matrix with best-matching unit (BMU) trajectory was generated. The BMU-matrix (50x1) of all trials of all players was exported and used for further quantitative analysis. Mean inter- and intra distance values were calculated with respect to the reference trial as was done by Bauer and Schöllhorn (4). Discussion and conclusions: Distance matrix Mean inter- and intra distance values à ratio (is there a difference between male and female players?) The next step could be to do a cluster analysis to this distance matrix Mean map quality parameters (QE and TE)? An interesting approach for a study of the validity of the map would be to use a Monte Carlo simulation with the input parameters of the neural network (initialization, training, learning rate and neighborhood radius) on their effect of the map quality parameters. References: (1) Kohonen (2001) Self-organizing maps (3rd Ed) Berlin: Springer-Verslag (2) Vesanto et al. (2000) The SOM Toolbox for Matlab 5 (3) Bauer and Schöllhorn (1997) Neural Processing Letters (4) Van den Tillaar & Cabri (2012) Journal of Sports Sciences