Gait Recognition Research Based On Skeleton Model And Sparse Depth Energy Image

As an important mean of biometrics,gait recognition has many prospects of applications in public safety,monitoring and detection.Especially with the development of Kinect devices,gait recognition has been brought into a new research boom.But at present,gait recognition is still a distance from a wide range of practical applications.How to get a design scheme with expansibility,strong robustness and good recognition effect is the key to the practical application of gait recognition.A series of studies,which are based on Kinect,are given in this paper.For the skeleton information,by constructing the 3D gait skeleton model and combining the method of human body partition with the gait skeleton model,the partly center of mass(PCM)and the total center of mass(TCM)are gotten to obtain the gait features which have high correlation with gait,such as the length of bone,angle and distance between PCM and TCM.For the depth information,active depth energy image(ADEI)is proposed which is based on gait depth energy image(GDEI).By combining the robust principal component analysis,a new sparse depth energy image method is proposed,which not only reflects the average information of the GDEI,but also contains the gait motion characteristics in the ADEI.By introducing collaborative representation to gait recognition,a new joint multi gait feature collaborative method is proposed,which can combine skeleton features with depth features.Then,the collaborative joint method of multi gait feature dictionary and the cooperative joint method of multi gait feature matching fraction are obtained respectively in the feature layer and the matching layer.After analysis and comparison,experiments show that the proposed method can effectively extract gait features and its recognition effect is better than other studies.In addition,the recognition effect of the collaborative joint method of multi gait feature dictionary is the best,whose average recognition rate is 97.87%.It's fully proved that the proposed method can reasonably combines multi gait features,which makes the gait features more effective and comprehensive,and the recognition effect is stable and robust under different walking conditions.