Research On Gait Recognition Based On Mobile Phone Sensors

Gait as a biological feature,with a stable,unique and not easy to forge features.Gait recognition plays an important role in identity verification,motion analysis and medical treatment.General gait recognition is based on image sequence to complete,and gait recognition based on gait acceleration sequence is a relatively new direction.Now with the rapid development of science and technology of the times,smart phones are often integrated with a variety of sensors.Therefore,how to combine these sensors and gait recognition technology to improve people's quality of life and safety factor has become a new problem.The acceleration data correction and preprocessing methods are improved.By combining the acceleration data obtained from the handset with the direction data and performing a vector decomposition operation.The collected acceleration data of the mobile phone coordinate system is transformed into the standard coordinate system of people's normal life through the improved formula.This improvement not only corrects the data,but also maximizes the intent of the characteristic of the acceleration data response in each direction.Then we use the db5 wavelet to decompose the corrected data from 2 to 3 layers,the data is denoised and smoothed,and the maximum periodicity of the data is reflected.Then five feature values including stride,pace,step frequency,standard deviation and kurtosis are extracted for gait recognition,and an improved algorithm based on multi-feature kNN classification is proposed.The values are assigned from the vertical and horizontal weights,and some ideas related to machine learning are borrowed.Finally,the correctness of step counting,steering judgment,motion state identification and identification is verified by experiments.The accuracy of the steering is close to 95%,while the accuracy of judging the static,walking and running is more than 90%,but the accuracy of the step-up and down-stairs is only about 80%.And 87.6% for the best accuracy of identity,4.8% higher than the traditional classification algorithm.Thus,multi-sensor gait recognition based on mobile phones is feasible,and also has a broad prospect,not only can record life bit by bit,but also to verify the identity to improve the safety factor.