Gait Recognition Based On Independent Component Analysis

Gait recognition tries to identify a person by the manner he walks. Compared with other kinds of biometrics such as face, iris, and fingerprint, gait has the merits of non-contact, unobtrusive, hard to disguise, and can be used for human recognition at a distance when other biometrics are obscured. It inspires the biometric recognition researchers working on it because of the increasing demand for automated human identification systems in security sensitive occasions.After principal component analysis (PCA), independent component analysis (ICA) is another useful tool for multidimensional data analyzing. ICA has shown its good classification performance in face recognition, but there's little ICA application to gait recognition. In this thesis, two new methods for gait recognition based on ICA are proposed. This thesis tries to find what factors affect gait recognition and to what extent.The first method is based on ICA Architecture I. Firstly, PCA is performed on image sequences of all persons and the PC axes are obtained. Then, ICA is performed on these PC axes to get the statistically independent IC axes. After that, the image sequences are projected onto these independent IC axes and the coefficients are obtained. The coefficients from the same person are averaged and the mean coefficients are used to be the representation of individual gait characteristics. For improving computational efficiency, a fast and robust method named InfoMax algorithm is used for calculating independent components. Gait recognition performance of the proposed method was evaluated by using USF gait dataset. Experiment results show the efficiency and advantages of the first method.The second method is based on ICA Architecture II. Firstly, PCA is performed on image sequences of all persons to get the uncorrelated PC coefficients. Then, ICA is performed on the PC coefficients to obtain the independent IC coefficients. The IC coefficients from the same person are averaged and the mean IC coefficients are used to represent individual gait characteristics. InfoMax algorithm is used for calculating independent components as in the first method. Gait recognition performance of the second method is evaluated by using CMU MoBo dataset and USF Challenge gait dataset. Experiment results show the efficiency and advantages of the second method.