Font Size: a A A

A Study And Design Of Invariant Feature Extraction Algorithm In Gait Recognition

Posted on:2018-07-17Degree:MasterType:Thesis
Country:ChinaCandidate:H F ChenFull Text:PDF
GTID:2348330536456283Subject:Pattern Recognition and Intelligent Systems
Abstract/Summary:PDF Full Text Request
The gait feature is a kind of biometrics,which has the characteristics of distance acquisition,non-contact and not easy to camouflage.Especially under surveillance,the distance between the camera and the pedestrian is far.The common feature like face,fingerprint and iris are not available,and gait feature which is pedestrian walking attitude is a more feasible method.Many researchers have demonstrated that gait recognition is a very effective method of identification at long distances.But gait recognition still faces various challenges such as multi-view,dressing change and carrying objects which lead to distortion and deformation of pedestrian gait contours,making it difficult to extract gait invariant features to distinguish between different pedestrians.So the current gait recognition can not be used in the actual scene.In view of multi-view,many researchers have proposed a view transformation model,which transform the gait feature from one perspective to another perspective.Most of the models need to estimate the perspective of the gait sequence,and a model can only be transformed between fixed two perspectives.This method can not effectively convert any view of the gait sequence to a specific perspective.At same time,it often needs other models to deal with other changes.We propose two gait invariant feature extraction models.One is SPAE method basing on multiple stacked auto-encoder,which is using the stacked multi-layer auto-encoder and progressive way to generate gait invariant features.The other is Gait GAN method basing on generative adversarial networks,which consists of three parts: generative model,real/fake-discriminator and correlation discriminator.The generative model converts the gait sequences under the influence of various factors such as different view to a gait sequence under a unified form.Real/fake-discriminator and correlation discriminator are used to constrain the gait sequences generated by the generative model.The advantage of these two methods is that it is possible to simultaneously solve the problem of multi-view,carrying objects and dressing changes in gait recognition without need to pre-estimate the state of gait sequence and only need one model to extract gait invariant features.The methods we presented are verified on both CASIA B and SZU RGB-D gait databases.The experimental results show that the two methods proposed in this paper can effectively solve the problem of gait deformation caused by the change of viewing angle,carrying items and wearing changes,and the extracted gait invariant features have good robustness.The recognition rate of the probe set has reached state-of-the-art and has a huge room for improvement.
Keywords/Search Tags:gait recognition, gait invariant feature, auto-encoder, generative adversarial Networks
PDF Full Text Request
Related items