With the development of computer vision and pattern recognition theory,gait recognition,as a novel biometrics,has attracted the attention of scholars in recent years.The goal of gait recognition is to identify individuals from the walking postures of the pedestrians.It has the features of long distance,and non-contact,hard to disguise,imitate or conceal and has a wide range of applications in areas such as video surveillance,access control and assisting detection in safety sensitive places.Since the walking direction of the pedestrian is arbitrary,the intersection angle between the walking direction and the optical axis of the fixed camera is also arbitrary,that is,the gait viewing angle is usually unknown.The appearances of the pedestrians can also be substantially altered.These two factors lead to the variations in the gait image of the same person from different gait perspectives to be larger than the variations in the gait image of the different persons from different gait perspectives and the recognition performance of the algorithm to drop sharply when viewing angle changes.In order to build a complete gait recognition algorithm with unknown viewing angles,improve the recognition performance and promote the development of gait recognition technology towards practical methods,a multi-view gait recognition algorithm based on subspace learning is presented in this thesis.The main contributions of this thesis are as follows:1)The gait viewing angle recognition algorithm based on the gait energy image projection information is proposed.Aiming at the unknown gait viewing angle problem,this thesis presents a gait viewing angle recognition algorithm based on the analysis of the characteristics of the gait energy image in different perspectives.The algorithm uses two-level gait view classification mechanism.First,gait view images are classified into two categories:front view images and rear view images,by using the gait energy image projection information.Then the viewing angles for both categories are identified respectively.The experimental results show that this algorithm can make full use of gait energy image projection information,improve gait viewing angle recognition algorithm performance,and achieve good recognition results.2)A multi-view discriminant analysis gait recognition algorithm based on view consistency is proposed.In order to improve the performance of multi-view gait recognition algorithm,this thesis proposes a multi-view discriminant analysis gait recognition algorithm based on view consistency.It uses the method of subspace learning,based on the MvDA framework and gait viewing angle recognition algorithm and applies the multi-view discriminant analysis based on the view consistency to the cross-view gait recognition.The experimental results show that this algorithm can optimize gait view discriminant analysis performance and improve multi-view gait recognition algorithm performance.3)A weighted fusion multi-view gait recognition algorithm based on view angle recognition is proposed.Aiming at inconsistencies between the view angle of gait images to be identified and the view angle of gait images in the test database,this thesis presents a weighted fusion multi-view gait recognition algorithm based on view angle recognition.This algorithm first implements gait viewing angle recognition,weighted fusion of near view angles.Then it takes the matching total score as the basis for gait recognition and identifies gait viewing angles not included in the database.Experimental results show that the algorithm is robust to the unknown viewing angle.In order to verify the recognition performance of the proposed algorithm,experiments were conducted on the CASIA-B data set of the Chinese Academy of Sciences which contains gait data from 11 viewing-angles of 124 people.74 people were randomly selected from the database as the training set,and the other 50 were used as the test set.The experimental results show that the accuracy of gait recognition algorithm proposed in this thesis reaches 86.18%when the recognition viewing angle is involved in the training,and 75.60%when the recognition viewing angle is not involved not involved in the training.The results show that the proposed algorithm is an effective gait recognition algorithm. |