Research On Tactile Gesture Recognition And Identity Authentication Based On Machine Learning

The widespread use of touch-screen smartphones has led to increasing access to sensitive and private data,which has led to the demand for security and available authentication technology.The purpose of identity authentication is to verify the real identity of the other end of the communication process and prevent counterfeiting and forgery.Biometrics-based identification technology is the current trend of identity authentication.It gradually replaces the traditional means of identity authentication and becomes a new type of identity authentication technology.Biometric recognition technology is based on the unique behavior and physiological characteristics of human body,through pattern recognition and image processing methods for identity recognition.Human biological characteristics have stability and uniqueness.Biometrics-based identification technology has broad application prospects and market potential due to its advantages of confidentiality,not easy to be forgotten,not easy to be stolen and so on.In biometric technology,the storage security of biometric information and the security of authentication algorithms are the keys of current research.In this paper,tactile gestures in biological features are mainly studied,and an appropriate identification algorithm is proposed.The system is implemented by a matrix pressure sensor.The paper first summarized the development status of the technology at home and abroad from the perspective of biometric technology used in identity authentication.Secondly,the stroke features of multi-user tactile gestures were analyzed,and the tactile gesture recognition technology was explored based on these features.We then studied machine learning algorithms and used a variety of machine learning algorithms to conduct a series of research experiments on Touchalytics tactile gesture dataset and UMD active identity authentication(UMDAA)dataset.Finally,the system was implemented and verified.The main innovations of this paper are as follows:(1)The Touchalytics dataset and UMDAA dataset were analyzed in the early stage.The datasets were preprocessed by MATLAB software to eliminate some erroneous data and reduce the impact of noise,so as to make the datasets more perfect.For each common type of touch operations,both static and dynamic features were extracted and analyzed for fine-grained characterization of users' touch behavior.The features were then combined for classification.(2)The improved extreme learning machine(ELM)algorithm is used to realize tactile gesture recognition and identity authentication for Touchalytics dataset and UMDAA dataset.(3)We used flexible matrix textile pressure sensor(based on Arduino platform)to collect seven types of tactile information from different users and verified the improved extreme learning machine(ELM)algorithm.