Accurate Recognition Of Keyboard Keystrokes Based On Sound Signals

With the development of acoustic localization and eavesdropping technology,keystroke detection has received tremendous attention from the academia and industry.With the development of smart phones,especially the various sensors on the mobile phones,mobile devices have been widely applied in localization problems to achieve centimeter-level accuracy.Although eavesdropping helps in designing keystrokes detection techniques,according to the principle of information security,a proper keystrokes detection mechanism can be applied to prevent eavesdropping further effectively.In general,all existing approaches can be classified into WiFi-based,vision-based and acoustic-based categories.WiFi-based methods use commercial Access Point(AP)and Network Interface Controllers(NICs)to detect gesture when one person keeps pressing on the keyboard.These methods mainly detect the keystrokes contact through the change of wireless channels.Vision-based methods adopt cameras and computer vision techniques to recognize keystrokes.The shortcoming of these methods is they are limited by light conditions.In order to solve the above problem,we propose a method based on Independent Component Analysis to detect the combined keystrokes.We conducted some experiments to verify that the previous method was not available in the study of combined keystrokes detection.Besides,we test the feasibility of Independent Component Analysis in the study of combined keystrokes.The main three contributions of this paper are listed as follows:Firstly,read the relevant papers to understand the research direction.In particular,we study the blind source signal analysis and independent component analysis techniques.Besides,we discuss the traditional method based on Wi Fi signal and acoustic localization technique.Secondly,design experiments,data collection and deploy experimental scenarios.In order to minimize noise due to factors such as environmental or accidental operation,we conduct the experiments in a quieter environment and take the noise reduction of sound signals.Thirdly,feature extraction,keystrokes detection.The FastICA algorithm is used to separate the mixed signal which is collected by the dual microphone.Besides,we utilize the machine learning method to classify and identify the keystroke content.The result of the experimentation has proved that smartphone equipped with our proposed scheme can detect combined keystrokes effectively.Furthermore,the average recognition accuracy of the proposed scheme is 78.4%.The main innovation of this paper is that we employ the Blind Source Separation for the combined keystroke detection.The system is simple.And it use a mobile phone,which is easy deployed and usability.