INTRODUCTION: Although underwater swimmers video recordings are widely diffused, swimmers` kinematics is usually reconstructed only on a single sagittal view, relying on manual digitization of specific features` position. In this study, 3D kinematics of the right arm in front crawl swimming is analyzed employing a markerless, automatic and repeatable method [1], based on multi-view video recordings of the subject. METHODS: An elite swimmer performing front crawl was acquired employing 5 subaqueous CCTV colour cameras and Canopus ADVC-55 A/D converters (PAL interlaced video, 25frames/sec). Cameras were synchronized automatically by means of an ad-hoc application, and calibrated with Bouguet method. Silhouette extraction was performed employing a Gaussian mixture algorithm (Intel OpenCV library), which creates an adaptive model of the background; a priori information, in terms of an extra "white" Gaussian component of the model, was included in order to deal with the presence of the foam. From the intersection of these silhouettes` back-projections in space, a visual hull of the subject was obtained at each frame. The joints` position was reconstructed by means of matching the visual hull with a subjectspecific mesh model (obtained from a dry and static visual hull of the subject), based on rigid-segments, employing the articulated-ICP algorithm. Only a manual initialization step is required, in which the initial position of the wrist is determined by digitizing its positions on all the views, and triangulating them. Correlation coefficients were evaluated for the three components of shoulder, elbow and wrist joints among three trials. RESULTS: The 3D joint trajectories repeatability analysis reported intra-method correlation coefficients between trials ranging from 0.948 to 0.997 for the antero-posterior component, from 0.751 to 0.998 for medio-lateral and from 0.2578 to 0.9975 for vertical. In the latter case, poor correlation was found for the elbow joint, while wrist and shoulder showed a good correlation (mean: 0.91±0.1SD). DISCUSSION: The application of a fully automatic markerless technique for analysis of swimmers` kinematics has been investigated. Good results were achieved in terms of intra-method repeatability of joint trajectories. Although this method has not been validated yet due to the lack of a gold standard, it is very promising for quantitative, wide-scale studies on swimmers` motion
© Copyright 2010 Biomechanics and Medicine in Swimming XI - Abstracts. Veröffentlicht von Norwegian School of Sport Sciences. Alle Rechte vorbehalten.