Font Size: a A A

Research On The Algorithm Of Identity Recognition Based On Gait Energy Image

Posted on:2018-08-17Degree:MasterType:Thesis
Country:ChinaCandidate:L Y JiaFull Text:PDF
GTID:2348330536959566Subject:Information and Communication Engineering
Abstract/Summary:PDF Full Text Request
In this paper,a novel spatio-temporal gait representationcombined with two improved feature extraction methods,which can significantly improve the gait recognition performance.Firstly,this paper introduces a newspatio-temporal gait representation,called gait energy image(GEI).The gait spatio-temporal representation using one gait image can not only preserves the movement information of human walking,but alsoretains the time information of gait.Compared with other methods of gait representation,it can significantly reduce the storage space and improve the real-time of gait recognition system,which provides a broad commercial space for identity authentication based on the gait.In addition,gait energy image is not sensitive to noise,which is conducive to gait recognition in real environment.And then,the gait energy image is combined with two improved image feature extraction methods.First,the improved phase congruency of gait energy image is extracted.The improved phase congruency algorithm is based on the improved local energy calculation method,frequency spread and noise compensation tactics.The feature of improved phase congruency is of good location and recognition.However,the global features of gait energy image have their limitations,such as the information of spatial distribution of the features extracted cannot be well reflected.Second,in order to overcome the limitations,wemerge the phase congruency feature of gait energy image with the improved spacepyramid SURF feature of gait energy image,so thatit not only retains the global feature of gait energy image,but also reflects the space distribution information of featuresof gait energy image.In this paper,an improved grid weighted algorithm,called partial least squares spatial pyramid representation(PlsSPR)is applied to the space pyramid of gait energy image to carry out the spatial pyramid weighted feature fusion.In this paper,the SURF features are extracted from each grid of all levels of gait energy image and the PlsSPR algorithm is followed before they are cascaded,which can well reflect the spatial distribution information of the feature.Finally,the gait features after fusion has serious redundancy,and the feature extracted from gait energy image is huge.While the existing gait database is relatively small.In order to avoid the small sample size problem and reduce the data redundancy,avoid the dimension disaster,and improve the processing speed,the improved PCA-LDA approach is used to decreasethe dimension of gait features and select the dominant features.To investigate the recognition performance of the proposed gait recognition method,the nearest neighbor classifier based on a novel ZCA whiting cosine similarity is tested on the CASIA DatasetB and USF gait databases.The experimental results show that our approach outperforms other state of the art automatic algorithms in terms of correct recognition accuracy and roubustness.
Keywords/Search Tags:Gait, Recognition Gait, Energy Image, Phase, Congruency PlsSPR, SURF, PCA-LDA, ZCA Whitening
PDF Full Text Request
Related items