Automatic parameter extraction methods for athlete swimming

Type: Semester thesis
Students: Dominik Breu, Jürg Germann
Advisor: Marc Bächlin, Kilian Förster

The purpose of this thesis is the investigation of automatic methods based on peak detection and template matching that extract significant parameters of the swimmers wrist acceleration.

Two automatic parameter extraction methods, the Peak detection and the Template matching, have been developed. Both methods recognized the different stroke events and types with a high success rate of over 90%. The Peak detection method performes better in the exact determination of turns. On the other hand the Template matching based algorithm copes better with fast changes of the swim stroke type.
A complete automatic method based on the Peak detection algorithm for the 7x50m Stroke Efficiency Test has been developed and compared with a manual measurement. The automatic method showed a comparable accuracy to the manual measurement. Figure 2 shows a processed 4x200m Individual Medley dataset with all detected stroke events and stroke types. Since both the Peak detection algorithm and the Template Matching algorithm do not require complex computations it is possible to implement them as onlinealgorithms on the SwimRecorder device itself. This would be a further step towards an improved and user friendly direct feedback system.