Gait Recognition based on Stochastic Switched Auto-regressive Model

A robust and compact gait model is desirable in many security applications because gait recognition is a promising non-intrusive biometric method. Only a few gait recognition systems adopted kinematical cues exclusively, but the dynamics model of parametric human body, including mass, length, inertia are seldom considered thoroughly. Furthermore, almost all these cues are velocity-dependent. The proposed model has a unique and flexible structure to deal with temporal features of gait like the timing and proportion of different phases in a gait cycle. It has a circular structure and 2 classes of states. In oder to fit the velocity-invariant features of gait, a special learning algorithm is proposed under the model's 2 kinds of structures. A 2-link virtual passive walking model plays an important role both in the configuration of the parameter matrix and the selection of the parameters' initial values. By evaluation the recognition rates of different models, the velocity-robust characteristics of the new model and its low computational load compared with conventional HMM are verified.