Abstract
In this paper, we have treated problem of persons identification by gait that is widely applied when people are far from the camera. To recognize pedestrians, we first propose a new markerless method based on a medical study to locate the interest body points from each silhouette. The proposed method is very suitable compared to existing works that use a harmonic predictive model which is sensitive to the speed variation. Then, we have introduced a new descriptor modeling the spatio/temporal variation of the silhouette's lower part that is considered among the most representative region in this context. To evaluate the robustness of the proposed descriptor, we have compared it with the most existing ones using either the Fourier transform, or the Hilbert transform applied on DWT transformation. Obtained results on CASIA (C) database clearly show that the proposed descriptor performances are very satisfactory and even surpass the very popular ones.