Research On Multi View Gait Recognition Based On Dynamic And Static Feature Fusion

In the field of intelligent monitoring and human behavior detection,compared to fingerprints,irises and other traditional biological featuresthe,the gait with the unique advantages of being non-contact,difficult to hide,difficult to imitate and disguise,and easy to collect,is a perceptible biological feature in the case of long distance.Therefore,the research on gait recognition technology has certain practical significance.The gait recognition process includes the preprocessing of gait video,extraction and expression of gait features and classification recognition.In order to obtain clear profile of moving target in the complex application scenes,based on the study of common target detection algorithms,the Double Difference Method with Gaussian mixture model achieves target detection,and acquires gait cycle and Gait Energy Image(GEI);Factors such as in backpacks and overcoats may affect the recognition effect when a single gait feature is identified,so,in order to reduce the impact of these unfavorable factors,taking the Gait Energy Image as static feature and the stride and gait frequency as dynamic feature,and they are fused by using a matching layer-based additive fusion algorithm after using the kernel principal component analysis with local preserving projection to reduce dimension of static features.Finally,in order to compensate for the shortage of single-view gait recognition,the fast decomposition orthogonal matching pursuit classification algorithm based on polynomial kernel function is used for classification and recognition combining the multi-view gait network.in addition,the classification recognition model is obtained through the process of testing and training.In this recognition model,the gait recognition rate in backpacks and overcoats is close to the gait recognition rate under normal walking state.The simulation results show that the Double Difference Method with Gaussian mixture model can detect the moving target and extract the clear moving target contour,and the effect is obviously higher than the traditional algorithm.To a certain extent,the fusion of gait dynamic and static features and the application of multi-view gait network improve the overall recognition rate with no affect of the interference factors such as backpacks and overcoats.In general,compared to the existing classical algorithms,the improved algorithm in the paper has certain advantages and good robustness.