Gait Recognition Method By Combining Body Contour And Its Region Bounded By Legs

With the technology of computer vision and pattern recognition developing, and the safety becoming more and more important, video surveillance system has been developing rapidly. Gait recognition, which uses the images and videos of pedestrians from computer vision identify a person, is one of the important researches.As a new biometric technology, gait recognition is mainly to identify a person by the pattern or style of walking. Gait recognition can identify a person in long-range and without contract, and it can be performed without interfering with the people. So gait recognition has been widespread concerned in the field of biological research. However, at present the research of gait recognition is still in theoretical exploration stage, and the environment and the complexity of human movement leads gait recognition becomes more difficult. Therefore, there are many problems in gait recognition, such as feature extraction, recognition rate and recognition speed. In order to solve the problems, the major tasks of this paper are as follows:(1) A comparative study of the extension methods based on LDA in gait recognition. In this paper a method, which is named a transformed PCA combined with RLDA method, are prospected. All the Eigenvectors correspond to nonnegative eigenvalues of the entire scatter matrix of training samples are selected to compose a lower-dimension transform space. In this transform space, in order to adjust the deviation and variance of the eigen values and overcome the small-sample-size problem, a regularization term is added to each sample class covariance matrix. And a new criterion function is established. By computing this optimization problem, and the eigen matrix is made up by some eigen vectors.(2) In order to solve the problem, gait recognition is easy to be interfered by the external environment, such as bag and coat, the region bound of legs(RBL) is found that they has significant discriminative information. So in order to improve the result of gait recognition, the gait recognition method based on the feature combination of gait image and its region bounded by legs is proposed. the features of gait image and its region bounded by legs are combined to represent gait features.(3) In order to improve the speed of gait recognition, the two-class Twins Fuzzy Support Vector Machine was combined with gait recognition. The SVM for two-class was transformed to a Multi-class SVM by pairwise classification. And the final result of classification of gait recognition are determined by voting.(4) In order to verify the effectiveness of proposed algorithm, the CASIA A,CASIA B and CASIA C, which are built by Chinese Academy of Sciences, are selected to use. And the two evaluation standard, which are recognition rate and recognition speed, are also used to measure the effectiveness of the algorithm. The minimum distance classifier, k-nearest neighbor classifier, Support vector machine and Fuzzy twin SVM are used to classify the pedestrians.