CMStatistics 2022: Start Registration
View Submission - CMStatistics
B1149
Title: Swimming technical skills tracking using multivariate functional clustering of Inertial Measurement Unit data Authors:  Antoine Bouvet - University Rennes 2 (France) [presenting]
Matthieu Marbac - CREST - ENSAI (France)
Salima El Kolei - ()
Nicolas Bideau - Laboratory Movement Sport Health M2S University Rennes 2 (France)
Abstract: Tracking technical skills during training becomes a major duty for coaches to improve the swimmer's performance. This can be done using miniaturized sensors such as Inertial Measurement Units (IMU). IMU data are multivariate functional data composed of six coordinates describing the 3D accelerometer and gyroscopic temporal records. To investigate the technical skills of the swimmers, two clusterings are performed based on the IMU records. The first clustering aims to provide groups with homogeneous swimming patterns that measure the efficiency of the swimming technique. This clustering is achieved by decomposing the IMU data into Fourier basis and thus fitting a mixture model on the functional basis coefficients. Since the number of basis coefficients is large, a variable selection is performed during clustering to select which coordinates of the functional data are discriminative. The second clustering aims to provide groups with homogeneous variations around the mean swimming pattern. It is performed on the residuals obtained by the decomposition of the IMU data into the functional basis. Thus, it aims to investigate how swimmers can reproduce their mean swimming pattern. We will discuss how the combination of both clusterings can be used to investigate the technical skills of the swimmers.