Human Gait Recognition And Application Based On Smart-phone

Biometric technology uses the inherent physiological or behavioral characteristics of the human body to identify individuals.The gait that the human body is difficult to hide and imitate is an important research area for biometric identification.Gait recognition through video capture and image analysis is costly and easily affected by clothing.Meanwhile,the collection and analysis through dedicated external sensors is difficult to promote due to additional purchases.In recent years,gait recognition based on smart phones,especially behavior identification methods,has been widely studied and put into use,but there is still a lack of research on identity recognition.This paper proposes a complete solution to the problem that gait recognition based on smart phones has no application in identity authentication.Gait data is collected by a smart phone,and then feature extraction is performed by using a dual-tree complex wavelet transform.Finally,a classification model is constructed by using machine learning techniques such as Random Forest and the identification is performed.This method is verified in a gait recognition system based on a client/server architecture.The specific work is as follows:(1)Data collection and feature extraction.Develop a software based on Android platform.The gait data of different ages and genders is collected based on sensors such as accelerometers in smart phones,and the differences in gait of different objects are analyzed.And then different data models are constructed according to the age and the gender of the objects.In view of the fact that the length limit of the actual filter can't fully satisfy many defects caused by the Hilbert transform conditions,the time-frequency characteristics of the gait data are extracted using the Q-shift modified dual-tree complex wavelet.And the parameter selection is analyzed and optimized.(2)Building a classification model.Weka is used as a tool for machine learning,with the number of instances of correct classifications and ratios,Kappa statistics,and confusion matrix as evaluation criteria.After analysis,Random Forest,Adaboost,K-NN and Bagging are used to build a classification model.And the knowledge flow tools are used to compare the performance of acceleration sensors and rotational vector sensors.The coordinate descent is used to conduct detailed research on the internal parameters of the four algorithms,and their impact effects are statistically analyzed.The optimal parameter scheme is selected to construct the classification model.Finally,by comparing different data models,it is shown that the age and the gender have an important influence on gait analysis.(3)Develop a gait recognition system based on Android client/servlet server for verification.The use of three-tier architecture and MVP,builders,factories and other design patterns for development to achieve "high cohesion and low coupling" software development requirements.Optimize the client in terms of interface layout,battery consumption and long time-consuming tasks.In terms of high concurrency and high consumption of the server,the asynchronous processing mechanism of Servlet 3.0,as well as the thread pool and database connection pool are optimized.An algorithm based on voting strategies is designed to vote in multiple models and multiple instances for final voting.The minimum value of the model and the algorithm are given when the identity recognition accuracy reaches 95%.