Analyse des signaux stabilométriques et de la stabilité chez l'Homme : application à la biométrie. (Stabilometric signals analysis and biometric application)

Biometrics refers to automatic recognition of individuals. It is based on their physiological and / or behavioral. The postural control, despite that is a human behavioral characteristic, has not been well developed in the field of biometrics. The work performed in this thesis is based on the stabilometric signals analysis ant biometric application. Firstly, a study of the postural information especially the stabilometric signal is carried out through traditional analysis namely temporal, frequency and stochastic analysis and two decomposition methods named principle components analysis (ACP) decomposition and wavelet decomposition. The ACP method, based on the additive model, allows decomposing the signal into three components: a trend signal, a rambling signal and a trembling signal. The wavelet decomposition method allows decomposing the signal into three levels of detail signals and three signal levels of approximation. Through the study of postural stability, spectral analysis and phase analysis of the different components from the ACP and the wavelet decomposition, the comparison of these two methods concludes that the ACP method is more appropriate than the wavelet decomposition to analyze the stabilogram. From the decomposition methods and classical methods of analysis, several parameters are extracted to study the effect of different factors on postural stability and the center of mass displacement. These factors are named vision, direction, proprioception, age, gender, height and weight. A second aspect of this work is devoted to the application of biometrics, from the extracted parameters and through ANOVA statistic analysis, those that are most discriminative are used to identify subjects and classify them according to age, gender, weight and size. This biometric application is performed by three classification methods namely, KNN, LDA and SVM. Biometric applications result in respectable recognition rate exceeding 80%. Therefore, it is inferred that the analysis of postural control is promising in the field of biometrics.