Research On Gait Recognition Method Based On Collaborative Representation

With the widespread application of the monitoring system, how to recognize the unidentified person has become to a topic that all people are concerned about. Gait recognition,aiming to identify a person by the way he walk, is favored by many researchers.Compared with other biometric features, gait has its unique advantages. It can be collected at remote distance and it is non-invasive. Also it is difficult to imitate and pretend. In addition,collaborative representation proposed in recent years has achieved better recognition results in the field of pattern recognition. Therefore, on the basis of collaborative representation, this paper study the gait recognition deeply.Firstly, aiming at the problem that single frame is sensitive to noise and can not reflect the motion characteristics of human, this paper uses gait energy image, which reflects the spatial and temporal characteristics of gait, as gait feature. To calculate the gait energy image,it is necessary to process gait images and test gait cycle. It also requires to standardize the silhouettes. Furthermore, in combination with the principal component analysis, the effective and low-dimensional features are extracted to reduce calculation.Secondly, with small samples of gait energy image of each object, the sparse representation method may lead to errors and time-consuming calculation. For this problem,we propose the method of gait recognition based on collaborative representation. It uses samples of all classes to represent the test sample and obtains the coefficients by using the regularized least square method. Then the test sample is classified to the class which gives the minimal regularized representation residual. Experimental results show that this method can achieve good recognition rate at a lower calculation time.Finally, we carry out the gait energy image data mapping from data space to high dimensional feature space by using kernel method. It can solve nonlinear problem and extract effective gait features. Because only using the linear character of gait in extraction process may produce wrong recognition results. Then, combining with collaborative representation,the final recognition results under different conditions are achieved. Experimental results on gait databases show that, although the view of gait image is changed, the proposed method performs certain robustness and improves gait recognition rate.