This paper presents a novel approach for extracting swimming performance parameters from accelerometer data using techniques traditionally applied to audio analysis. The recorded acceleration data is treated as sampled audio data, with the stroke rate (one of the main parameters to extract) treated as the fundamental frequency. A pitch detection algorithm is then adapted to this domain and applied to the data. Lap counts are derived from an analysis of the output frequency series as percussive events, and stroke counts are obtained from a pitch-aligned adaptive spectral extraction algorithm. This approach was compared to conventional accelerometer data analysis techniques (peak detection and zero crossing) and demonstrated superior accuracy compared to a visual observation criterion. For stroke count determination, the new approach had zero error rates of 86.1 % (versus 58.3 % for the zero-crossing method) and 75 % (versus 60 % for peak detection) for lower-back and wrist placement locations, respectively. Additionally, this new approach is not dependent on filtering, and is robust to sensor displacement and axial alignment.
© Copyright 2015 Sports Engineering. The Faculty of Health & Wellbeing, Sheffield Hallam University. Alle Rechte vorbehalten.