IDENTIFYING INDIVIDUAL MOVEMENT STYLES IN HIGH PERFORMANCE SPORTS BY MEANS OF SELF-ORGANIZING KOHONEN MAPS

  • Wolfgang I. Schöllhorn
  • Hans Ulrich Bauer
Keywords: pattern recognition, individuality, kinematography, javelin throwing

Abstract

INTRODUCTION: Although single case studies are common in more general oriented behavioral sciences (Yin 1988), they are still very rare in sports science. The still lasting preference of group studies in sports science is often connected with the assumption of ‚ideal techniques‘ in the investigated sports disciplin. Wether such an assumption is justifiable biomechanically, should have been investigated in this pilot-study by means of a pattern recognition approach. METHODS: The final throwing phase of 8 male and 19 female javelin throwers was filmed with two highspeed cameras threedimensionally. The male throwers were finalists of the world championship 1987 in Rome, whereas the female group consisted of 10 worldclass heptathletes and 19 javelin specialists with national and international level. From 2 female specialists 10 and 6 throwing trials were filmed in different competitions, respectively. The throwers movements were described physically complete by means of the main joint angles and its angular velocities. Kohonen maps were trained to project these highdimensional individual feature vectors to a low dimensional neuron output space. The euclidian distances between all trajectories in the neuron output space formed the basis for a cluster analysis (Bauer/Schöllhorn 1997). Beside the complete set of variables we seperated the movement into variables of the a) lower and upper body, b) left and right side, and c) angles and angular velocities. RESULTS AND DISCUSSION: The cluster analyses of all variable sets do not display any performance dependance. Clusters of male and female techniques are distinguishable only tendentious in the complete variable group. What we could identify in all sets of variables were the 10 and 6 trials of the two female specialists, respectively, within two seperated clusters. Although both athletes‘ throws had similar thrown distances (55m to 68m), they were not in the same cluster, but were seperated from the others completely. These clusters, even in the subgroups of variables, provide a clear indication for highly individual throwing techniques not only in the whole but also in upper and lower bodies movement as well as in the left and right side movement, and in the angle and angular velocity, respectively. CONCLUSION: The identification of individual throwing styles by means of a duration of 200ms leads to rethink the idea of ideal throwing techniques and its pure imitation in learning strategies. REFERENCES: Bauer, H. U. Schöllhorn, W. (1997). Self-Organizing Maps for the Analysis of Complex Movement Patterns. Neural Processing Letters 5, 193-199. Yin, R. K. (1988). Case Study Research. Newbury Park: Sage Publications. * We would like to acknowledge Dr.Menzel for providing the data.