Research On Gait Recognition Algorithm Via Random Subspace Method

Recognition by gait is a new and attractive research field for the computer vision andbiometrics recognition technology. Its aim is to recognize people or detect physiological,pathological and mental characteristics by their walking style. Compared with otherbiometrics recognition technologies, such as fingerprint, iris and face, gait is human’s explicitand dynamic representation which is closely related to the information of spatial-temporalwaking movement. Gait recognition is the unique perceivable biometrics at a distance and hasthe advantages of being noninvasive,requiring little about the quality of video and difficult todisguise. It is an attractive biometrics recognition technology in the field of modern nationaldefense construction,terrorist attacks and security.The key point of gaitrecognition is to extract suitable static or dynamic features or theircombination when analyzing the human walking videos captured from cameras. Gaitsignatures are then derived and integrated with classifiers to achieve gait recognition. Itcontains many kinds of techniques, such as computer vision, pattern recognition, video andimage sequences processing and so on.Based on its strong points and importance both in theoretical research and practicalapplication, gait recognition will be further studied based on the deterministic learning theoryin this dissertation. The main contribution and innovation of this dissertation are summarizedas follows:Firstly, investigate the pre-process of gait recognition. The first to use the Gaussianmixture model and the difference method to the extraction of the silhouette of the input video,and then compare the detection method of the gait cycle.At last, different gait database arecompared and the gait energy image templating the head portion center of gravity forregistration which accurately extract the characteristic data.Secondly, propose gait recognition algorithm based on principal component analysis(PCA) feature extraction and dimension reduction, using support vector machine (SVM) as aclassifier. By experiments, the shooting angle is fixed, the same type of clothing, compared tothe traditional algorithm in the recognition rate and speed performance enhancements lot. And compared the performance of the algorithm under11shooting angle on the use of statisticalmethods for the best viewing angle.Thirdly, For real life based on the fact that clothing are differrnt, a gait recognitionalgorithm based on stochastic subspace is proposed. First, for the PCA to overcome the curseof dimensionality, introduced2DPCA and on this basis to be extended to E2DPCA, C2DPCAand W2DPCA of feature extraction and dimensionality reduction; then in order to avoid theclassifier over-fitting clothing type, the introduction of random the idea of sub-space, andcombined with the principle of voting to reduce the error caused by randomness. Theexperimental results show that, compared with the traditional algorithm, the method used inthis paper on the recognition rate up to25%-30%.Fourthly, The stochastic subspace dimension and the number of optimization problem:when the subspace number increases, the recognition rate unlimited approaching a certainthreshold; when changes in the dimension of subspace, the classifier’s state changes fromunder fitting to over fitting.Finally, investigate effects of types of clothing and body parts for gait recognition:through the study of different types of clothing and body parts, find out that the type ofclothing to cover most of the body difficulty identifying details, to use the local physicalcharacteristics can effectively improve the recognition rate.