Algorithmic Activity Detection Based on Motion and Position Sensors

This thesis deals with the segmentation of repetitive sports exercises. The segmentation of such time series data constitutes a scientifically interesting problem, as it enables analysis of individual repetitions using techniques such as machine learning. To determine whether it is possible to implement a segmentation algorithm that can run on a micro-processor unit, this thesis proposes a prototypical design. An experiment was conducted in which test persons performed exercises while equipped with sensors. A segmentation algorithm was devised and subsequently used to process the signal data collected. The segmentation was evaluated by comparing the algorithmically determined segmentation points with a ground truth based on video recordings. The algorithm exhibited linear time complexity and a low failure rate. A window size of 501 data points resulted in 0.02% false negative segmentation points and 0.04% false positive segmentation points. The study shows that a suitable algorithm can be implemented.