Gait Recogniton Reaserch Based On Feature Subspace

Gait recognition is a emerging field of biometrics recognition research and is aidentity recognition technology by the walking pattern of human being. Compared withother biometrics, gait feature is the only biometric that can be perceived at a distance alsohas the characteristics of non-contact, low requirement for pixel resolution and highdifficulty of concealment and disguise, so gait recognition has a lot of potential practicalsignificance and application value for smart surveillance and safety protection.Although a series of research achievements has been made, gait recognition is still inthe studying and exploring stage. Compared with the fingerprint, iris and face recognition,its accuracy is low and lack of robustness. This paper focus in gait feature’s extracting,aimed at improving the recognition accuracy. Its main work is as follow：Firstly, a gait recognition method on the gait energy iamge feature fusion is studied.Gait energy image is selected as the original feature, using the number of frames of thegait sequences instead of gait cycle to compute training set’s gait energy images. And thenits histogram of oriented gradient, wavelet and local information entropy features areextracted; Principal component and canonical correlation analysis are used to reduce thedimension of training and experiment sets’ features; gait features are fused and with thehelp of classic classification algorithm, the training and experiment sets’ gait features areclassified, computing the gait recognition rate. HumanID Gait Challenge data base’s gaitdata is used to do experiment and result analysis..Secondly, Chrono-gait image feature and cosine and norm distance similaritymeasure are studied. Chrono-gait image is extracted and disposed by means of color spacetransformation method.1-Nearest Neighbour algotithm with l1-norm similarity measure isused to compute gait energy imge and RGB, YCBCR color space’s chrono-gait image’sgait recognition rate respectively and compare gait feature’s recognition performance;YCBCR color space’s chrono-gait image is used as classification feature，computing itsgait recognition rate in the1Nearest Neighbour claasification algorithm with cosine andnorm distance similarity measure used in gait recognition and compare classification effect. HumanID Gait Challenge data base’s gait data is used to do comparison experimentand result analysis.Finally, a optimal feature fusion method is studied. Choosing two features which canobtain highest gait recognition rate from all the original ones as the optimal features andreducing their dimensions. Then we use canonical correlation analysis with addedregularizatioan term to get their canonical correlation features. After feature fusion, by1-Nearest Neighbour algorithm with cosine similarity measure, the gait recognition rate iscomputed. HumanID Gait Challenge data base’s gait data is used to do experiment andresult analysis.