The aim of this study was to identify kinematic and dynamic variables related to the best tumble turn times (3mRTT, the turn time from 3-m in to 3-m out, independent variable) in ten elite male swimmers using a three-dimensional (3D) underwater analysis protocol and the Lasso (least absolute shrinkage and selection operator) as statistical method. For each swimmer, the best-time turn was analyzed with five stationary and synchronized underwater cameras. The 3D reconstruction was performed using the Direct Linear Transformation algorithm. An underwater piezoelectric 3D force platform completed the set-up to compute dynamic variables. Data were smoothed by the Savitzky-Golay filtering method. Three variables were considered relevant in the best Lasso model (3mRTT=2.58-0.425 RD+0.204 VPe+0.0046 TD): the head-wall distance where rotation starts (RD), the horizontal speed at the force peak (VPe), and the 3D length of the path covered during the turn (TD). Furthermore, bivariate analysis showed that upper body (CUBei) and lower limb extension indexes at first contact (CLLei) were also linked to the turn time (r=-0.65 and p<0.05 for both variables). Thus the best turn times were associated with a long RD, slower VPe and reduced TD. By an early transverse rotation, male elite swimmers reach the wall with a slightly flexed posture that results in fast extension. These swimmers opt for a movement that is oriented forward and they focus on reducing the distance covered.

Human action recognition in video is one of the key problems in visual data interpretation. Despite intensive research, the recognition of actions with low inter-class variability remains a challenge. This paper presents a new Siamese Spatio-Temporal Convolutional neural network (SSTC) for this purpose. When applied to table tennis, it is possible to detect and recognize 20 table tennis strokes. The model has been trained on a specific dataset, TTStroke-21, recorded in natural condition (markerless) at the Faculty of Sports of the University of Bordeaux. Our model takes as inputs a RGB image sequence and its computed Optical Flow. After 3 spatio-temporal convolutions, data are fused in a fully connected layer of a proposed siamese network architecture. Our method reaches an accuracy of 91.4% against 43.1% for our baseline.

No abstract