Biomimetic gait recognition based on motion contours wavelets analysis and mutual information

A novel gait expression and recognition algorithm is introduced here. After the wavelet transforming, a Spatial-Temporal Energy (STE) images from one gait sequence yield an approximation sub-image including the body´s main expression, and three detail sub-images including the motion feature. With the mutual information the combination feature is got. The constructed combined gait feature made use of the microscopically analysis and orientation electivity of the wavelet transformation, which could reconstructed the lossless gait data. And those coefficients for the sub-images would be determined with the maximum mutual information made those discriminative features more outstanding. Hence the gait recognition is more similar to the mode of the human visual. This expression combined the static and the dynamic information of gait, which used selective judgement of the human to the motion for reference. And it deals with the problem of time-dependence and the gait cycle extraction. The proposed method is evaluated on the SOTON dataset, and the performance is compared with some algorithms. Experimental results demonstrate that the combined feature has the universality. And the method has higher recognition rate, as well as the merit of low space and time requirement.