Research And Design Of Gait Recognition Based On Smart Phone

Nowadays,as COVID-19 continues to ravage the world,and people are becoming infected or dying from the disease almost every day,how to prepare for the epidemic has become the focus of attention.Avoiding contact with others by wearing a mask is the most direct way to reduce the risk of infection,however in some public places it is necessary to remove the mask for user identification.The focus is on how to identify people while wearing masks.Therefore,this thesis mainly studies the smart phone-based gait recognition method and design,and realizes a set of realtime identity recognition system,its main work is as follows:Firstly,this thesis presents a method of data acquisition and processing.The method mainly includes pre-design data acquisition scheme,gait signal selection,data acquisition and data preprocessing.In the selection of gait signal,through experimental analysis,it is found that the combination of acceleration and gyroscope acquisition of gait data can more accurately identify the identity of the subject.In terms of data acquisition and processing,this thesis on the mobile end using acceleration sensor and gyro sensor designed the data acquisition module,real time collection and the subjects’ gait data and uploaded to the server,and the PC to preprocess the data collected,in order to reduce the noise influence on experimental results,eventually improve the identification accuracy.Then,a method of feature extraction and identity recognition is proposed.Firstly,gait data of30 subjects at hand,waist,ankle,arm and wrist were collected in the early stage of the experiment,and NJUPT data set was established.Through experimental comparison,it is found that the data set of waist position has the best performance in the experiment,while the data set of hand position has not so ideal performance.However,placing smart phone in hand is the most common and convenient way to collect data.Therefore,when the data set was expanded in the later stage of the experiment,gait data of the waist and hand positions of 83 subjects were collected in this thesis,and the previous 30 subjects were combined to form a NJUPT data set containing 113 subjects’ gait data.Secondly,this thesis discusses the influence of sliding window size on accuracy by using the data set of hand position.Then,in the aspect of feature extraction,this thesis studies the time domain features which are easy to calculate.Through experiments,it is concluded that the time domain features such as mean,standard deviation,median,maximum and minimum are more suitable for classification.Finally,when in the identification model,this thesis comparative analysis of the decision tree,random forests,K-nearest neighbour,support vector machine(SVM),naive bayes and XGBoost performance and operation time,random forests is chosen as the final final classification model,and on the tuning parameters,the best identification model is set up,will eventually deploy the model to the mobile terminal.The results show that 94.05% accuracy can be achieved by using the data set in hand position.Finally,a real-time identification system is designed and implemented based on smart phones.It mainly includes the overall design,identity module design,database design and system test.In the overall design part,it mainly introduces the overall architecture,development environment and mobile terminal functions.In the identity module part,it mainly introduces the framework and process of identity identification.In the database design part,it mainly introduces the table design for storing data.In the system testing part,the identification system is tested.