THE USE OF KOHONEN FEATURE MAPS IN THE KINEMATIC ANALYSIS OF ROWING PERFORMANCE

Abstract

This study used the self organising map (SOM) as a processing step in reducing the complexity of human movement dynamics obtained from execution of a rowing stroke. The SOM is an artificial neural network (ANN), adapted in an unsupervised manner using a self organising learning process. Three-dimensional joint angles, produced by the rowers, were projected onto a 2 dimensional topological neural map, thereby identifying rowing movement patterns. Unsupervised clustering allowed the time series rowing strokes to be positioned on the map in relation to each other and this enabled movement patterns to be compared. The larger kinematic variation of the novice rowers was observed. The weight vector associated with each SOM cluster illustrated underlying task related changes in the rowing stroke patterns between elite and novice.