LLE based gait recognition

LLE based human gait recognition method is proposed. LLE representation is based on the similarity of images and is inherent translation, rotation and scale invariant, so LLE representation of gait sequences can be viewed as a kind of dynamic features of gait. 1D LLE representation based gait cycle detection method gives complete gait cycle and is invariant to visual angle, therefore it is better than classical silhouette width or silhouette area based methods. Dimensionality of LLE representation is related to the DOF of gait and 10D LLE representation of simple gait is sufficient for gait recognition. The recognition rate is 100% on CMU MOBO database. If static features are combined with dynamic features, perfect recognition rate will be achieved.