Technology And Research Of Implicit Identity Authentication In Mobile Devices Based On Gait

With the spread of mobile Internet and the development of mobile devices,people's Daily life is increasingly dependent on smart mobile devices.While enjoying the convenience and efficiency brought by mobile APPs/services,individuals produce an army of data or using traces.These data and traces contain users' behavior information that can be proposed for the individual and distinguishing authentication features.And we can provide better authentication services to mobile devices and applications base the features.In this paper,we pay attention to the mobility of mobile devices and collected the acceleration sequence which generated from users' walking with the devices.Furthermore,we extracted the authentication features from these data to achieve the behavior-based,non-intrusive authentication system.This paper is driven by the essence of authentication and the current mainstream biometric authentication scheme,which proposes following tow completely different authentication algorithms,and these two different algorithms have obtained meaningful results in the real data experiments.The first method is that gait acceleration sequence authentication algorithm based on the classification.The acceleration sequences generated by users' walking are composed of an army of similar subsequence step(cyclic sequence)together.So we extract the each user's step cycle sequences,and transform the step sequence cycle into the frequency-domain space to descript the users' features.And finally,we use these features to construct classifiers for each user as a user authentication model.The second method is that the gait acceleration sequence authentication algorithm based on the self-regressive.We found that the method based on classification could not meet the real-time capability,and the method based on classification easy to raise the unbalanced learning problem,so we proposed another method based on self-regression.We regarded the gait acceleration sequence as a time series based on autocorrelation.And we constructed each autoregressive model for each user as the authentication model.Theoretically,this method avoids the problem of unbalanced learning and should obtain better performance than the method based on classification.In order to improve the fitting capacity of the autoregressive model,we propose a new recurrent neural network model,SNTM,which is more suitable for this problem,and we evaluated the SNTM has better performance in our experiment.In this paper,the feasibility of the two methods is verified from the actual data.In general,this paper has contributed to the research field of biological authentication and recognition.