The Research On Gait Recognition

As enhance of the security requirement in the modern society, Biometrics Recognition, as an approach for society security, has been greatly researched and developed. Early psychological studies into gait suggested that gait was a unique personal characteristic. People can identify persons by observing their walking. Gait Recognition research involves in following areas: movement segmentation, feature extraction, pattern classification and gait database and so on. In this paper, the fore three areas are mainly researched.On the movement segmentation, an adaptive background subtraction algorithm is used to subtract gait sequences background. Then the gait image subtracts the background image to extract body area. Binarization and morphologic algorithm are used to remove image noises. An approach based on human body model has been proposed to remove body shadows. This approach gains precision result of the gait.On the feature extraction, an approach based on Hu invariant moments to extract gait features is proposed. Then it is modified to the other approach based on Zernike moments to extract gait features. All of gait Zernike moments compose the gait feature space. Because the space has high dimension, PCA algorithm is used to compress Zernike moments and a new feature space containing gait temporal features can be generated. Then Zernike moments gained from gait image are projected to the new feature space and a new vector can be gained. All of vectors gained from a gait cycle compose a feature matrix. Finally the row variance vectors considered as classification features are computed from the matrixes.On the pattern classification, we bring out a new concept called Normalized Gait Cycle. We consider it as based cell in gait classification and find the approach to compute it. BP neural network is used to classify the gait features.Finally, all kinds of the approaches are realized by programming. We present a Gait Recognition System developed by ourselves. It is a platform for new algorithm research and test. When we use Little&Boyd gait database to test our system and approaches, 100% recognition rate is reached.