A Research Of Inertial Sensor-Based Gait Identification

The popularity of smart devices and human's concern of privacy data security have prompted people to find new long-term identity recognition mechanisms.The current popular fingerprint and face biometric technology can only provide one verification.That means when the device is in the "on" state it will not be able to verify who is its real owner.And such biometric technology requires close contact with the device,it will not work in non-contact or long distance occasions.Medical and psychological research shows that human gait is unique and it's hard to imitate,it can realize long-distance,non-contact,implicit identification of user identity.Inertial sensor-based gait recognition is a hot topic in current academic circles.This study uses the built-in acceleration sensor of the smartphone to obtain acceleration data for gait recognition,focusing on solving the problem of uncertainty of placement of smartphones in practical application environment.The specific work is as follows:(1)Data collection and pre-processing.This paper first obtains the built-in accelerometer of the mobile phone,develops its own data acquisition software.Then a data collection scheme was developed,12 volunteers put the smart phone on the waist,front pocket and hand walking normally to collect original gait data.Then,a series of preprocessing operations such as acceleration signal synthesis,filtering,smoothing,dividing gait cycle,and removing the abnormal cycle are performed.Finally,feature extraction in the time domain and frequency domain is performed for each gait cycle.(2)Study the algorithm model and parameters of personal identity authentication.The article selects the universal algorithm SVM to experiment,and selects two common kernel functions of Gaussian kernel and polynomial kernel of SVM model.Gaussian kernel width was set to 1,the polynomial kernel order was set to 3.Finally a series of model index comparison results prove that the Gaussian kernel of the SVM model is efficient in solving this problem.(3)Research on multi-person identification algorithm model and parameters of adaptive mobile phone placement.For the gait multi-person identification work mode,this paper uses the random forest and KNN,SVM,Bayes network,BP neural network for comparative analysis,the parameters of each model are studied,and the hybrid model is considered regardless of the location of the mobile phone.The gait identification is verified,and the best identification model for this study is random forest,and the best results is got when the number of trees is set to 50,and the number of candidate feature subsets is set to 6.However,when the best random forest is used on unrelated test set,the average accuracy got by the hybrid gait recognition model is only 82.19%.Then the gait recognition under the hybrid model is optimized,and the model of merging the position information of the sensor is proposed,which is the hierarchical identification scheme using position + identity.The location recognition is implemented by the decision tree,the identity identification is realized by random forest.The accuracy of the pocket,hand and waist parts of the optimized solution is 99.9%,98.3% and 97%,respectively,and finally reaches an average of 91.95% on the irrelevant test set.In general,the optimized program not only improves the accuracy but also adapts to the location of the phone.(4)Develop an identity prototype system which is adaptive to mobile phone placement.The system integrates data collection and identification functions.Firstly,We designs the client UI interface based on the Android platform,and implements the Java language encoding for data acquisition,preprocessing and feature extraction module.Secondly,weka is used to built in the location and identify model to mobile phone.Finally,after the user collects unknown gait data,this system will give the final recognition result.