Analysis of radar human gait signatures

The authors develop methods for the time-frequency (TF) analysis of human gait radar signals. In particular the authors demonstrate how knowledge of different motion classes can be obtained via a Markov chain model of state transitions based on the TF envelope structure associated with the motion sequence being analysed. The class-conditional knowledge thus obtained allows us to effectively extract the motion curves associated with different body parts via a non-parametric partial tracking algorithm that is coupled with an optimum Gaussian g-Snake modelling of the TF structure. The optimum segmentation of the TF structure into different half-cycles as well as the determination of the initial Doppler control points (corresponding to each half-cycle) is facilitated by a dynamic programming algorithm wherein the associated cost function involves a mean-square minimisation of the best quadratic fit to each segment together with a sparsity prior that enables us to control the smoothness of the approximation space in which the time series being analysed is effectively projected. Finally, the authors describe some of the limitations of our approach and point out future research directions that can overcome some of the difficulties associated with the complex interaction between the inherently non-linear dynamics of human gait motion and radar systems.