User Identification Based On Mobile Phone Sensors

Smart phones have become an indispensable part of people's production activities and life.The embedded sensors have the characteristics like miniaturization,multifunction and intelligence.Smartphones are powerful,compact and portable,but they are also easy to lose.A common way to prevent loss is to install specific software on the mobile terminal,which enables users to locate and control the mobile phone remotely after it is lost.This is a method of post-perception,which can't be remedied when the user realizes that the mobile phone is lost.This thesis uses human walking characteristics to identify users.This kind of user recognition does not require direct human-computer interaction and has no special requirements for the environment.It can minimize the loss of mobile phone to users.The methods of identifying users in smartphones mainly include independent password,fingerprint recognition,face recognition,etc.But these methods cannot help mobile phones identify their "masters" without direct human-computer interaction.The main work of this thesis is to realize user identification by analyzing the data collected by mobile acceleration sensor and gyroscope.Acceleration sensors can acquire the acceleration in the three-dimensional direction of the mobile phone,and gyroscopes can acquire the three-dimensional angular velocity when the mobile phone deflects.These data contain the unique behavior characteristics of users.Nowadays,Deep learning has made remarkable achievements in many fields,so it is feasible to use Deep learning to complete the work of this thesis.The main goal of this thesis is to use deep neural network extract the unique features of users from the data taken by mobile phone sensors to realize the function of user recognition.In this thesis,we focus on the acquisition of mobile sensor data,data preprocessing,data set generation,deep neural network structure and user identification.In the whole experiment process of user identification,the accelerometer and gyroscope are used to collect data,and then the collected data are preprocessed to generate their own data sets and input into the deep neural network for feature extraction and classification.In the construction of deep neural network framework,firstly,the data collected by different sensors are input into the independent convolution neural network to extract their own data features,and then the two features are fused and then input into the fusion convolution neural network layer and the self-attention mechanism layer in turn to extract the external relations and temporal features of the two data.Finally,these features are input into the full connection layer to classify and output the classification results.This thesis introduces several experimental processes and results of user identification under different depth neural network frameworks.From the experimental results of these comparative experiments,it can be seen that the deep neural network frameworks described above perform best in the experiment,and the accuracy rate reaches 97.5%.This thesis mainly studies on the premise of closed-set,which belongs to the preliminary work.In the follow-up,the network will be applied to the open-set to achieve user identification.