Homological Analysis For User Identification Based On MEMS Sensors Data

In recent years,the issue of smartphone security has attracted much attention.User identification plays an important role in ensuring the security of smartphones.With the increasing requirements for the security of portable devices,various user identity authentication technologies and tasks have been proposed for smart phones,including identity authentication based on accelerometer data.However,the time series based on accelerometer data collection are very complex with highly non-stationary characteristics.The statistical characteristics of the signal vary greatly over time,which brings challenges to user ident ification research.In this work,accelerometer signals collected from ten volunteers during walking,walking downstairs and walking upstairs are adopted in our experiments.A new online user authentication model is constructed by homological analysis of gait data.Instead of the traditional authentication method based on time domain and frequency domain characteristics,an online user identification method based on Micro-Electro-Mechanical Systems sensor data is proposed by homological analysis of gait data.The main research work includes:1.The statistical characteristics of gait time series signals based on time domain and frequency domain and the topological characteristics based on phase space are described in detail in this thesis.For the construction of the topological characteristics of the phase space,firstly,the acceleration signal is mapped to the phase space,and continuous coherence is used to describe the topological structure of the acceleration signal in the phase space.Secondly,use Persistence Diagrams(PD)to describe the topological characteristics of single-segment gait signals,derive and calculate the probability distribution of PD.Finally,in order to reduce the randomness that the uncertainty caused by noise brings to the topology of the PD,a feature that is more robust than a single PD is constructed,that is the Expected Persistence Diagrams(EPD).Because different people have different walking styles and habits,the topological structure and topological characteristics of each person's gait time series in the phase space will also be different,resulting in the EPD constructed by the conversion of different people's gait time series.For the difference,the inherent walking patterns in the accelerometer data of different users.2.A user authentication system based on Support Vector Machine(SVM)and a user authentication system based on Probabilistic Neural Network(PNN)are proposed.With SVM as the classifier,the accuracy of identity authentication based on gait signal statistics is 82.68%;with PNN as the classifier,the accuracy of identity authentication based on gait signal statistics is 85.5%.At the same time,the EPD-based features are input into SVM and PNN,and the results are better than those based on statistics.With EPD as the feature,the SVM-based identity authentication accuracy rate is 92.73%,and the PNN-based identity authentication accuracy rate is 92.08%.3.An user authentication system based on Kullback-Leibler divergence(KLD)is Proposed.The difference in walking patterns of different users can be measured by calculating the KLD of the EPD distribution function.The collected unknown user data can be calculated by calculating the KLD with the potential model generated by the user's pre-stored training data to determine whether it is a matching user Based on the features of EPD and the decision measurement of KLD,a robust online user recognition system is constructed,and the recognition accuracy based on KLD reaches 95%.