Intrinsic mode functions for gait recognition

Gait recognition is an attractive biometric as it is unobtrusive and can be used for recognition from a distance. A number of methods have been proposed by different researchers in the recent past for this purpose. Most of these methods analyze gait as a linear and stationary signal. However, recent research shows that gait is nonlinear and non-stationary. Hence, linear analysis would be insufficient for analysis of gait. In this paper, we present a novel recognition algorithm that derives the feature vector by performing nonlinear, non-stationary analysis of gait using a technique called empirical mode decomposition. We test the algorithm with both noise-free and noisy gait sequences and demonstrate its applicability.