Research On High Precision Acquisition Technology Assisted Identification Method Based On Footstep Signal

With the rapid development of big data,Internet,computer technology,data acquisition methods and equipment,personal identification technology have been used in national public safety and supervision services,perimeter security,military reconnaissance and health care.Traditional identification technology is mainly based on identity card,password,smart card and other technologies.These traditional identification methods usually have disadvantages such as inconvenient,easy to be stolen,forged,lost and so on.It is difficult to meet the needs of the rapid development of modern social science and technology.In recent years,biometrics using iris,fingerprint,face,gait,voiceprint have emerged.It utilizes the relatively stable physiological characteristics of the human body that has certain physiological basis,can reflect the behavioral characteristics of human psychological changes.The identification of personal identity can be realized by combining with computer technology such as image processing and pattern recognition.For example,iris recognition can identify a user's identity by comparing iris image features.Fingerprint identification can use the unique characteristics of fingerprint to collect and store images and texts,and identify the user's identity through different fingerprint patterns and lines.Face recognition that can identify personnel by collecting and analyzing face image or video from camera is more widely used.Voiceprint recognition is also a type of biometric technology,also known as speaker recognition which performed by causing differences in the size and shape of human speech organs and analyzing the voiceprint.Gait recognition is the acquisition of walking video through a camera to extract features from the gait contour,so as to realize identity recognition.Although the biometric technology introduced above has been well applied and developed,biometric recognition techniques based on iris,fingerprint,and voiceprint usually require active cooperation of participants or users.Biometric technology based on face and gait needs to analyze the collected images or videos.For some areas where it is inconvenient to install cameras,such as perimeter,private places,secret areas,etc.,the two technologies will be limited.For some illegal and criminal activities,suspects will deliberately avoid the area with cameras and engage in illegal and criminal activities such as smuggling,stealing and escaping that can not well monitor through iris,fingerprint,face and other biometric technologies.Therefore,this paper studies a biometric technology based on footstep vibration signal belongs to a kind of human behavior that contains the personality,gender,age,height and weight of the individual.The identification of personal signals can be achieved by analyzing the footstep vibration signals.This paper mainly studies related aspects including design and development of high precision acquisition instrument,signal noise reduction,feature extraction,feature selection,classification and identification.(1)Firstly,the mechanism of the footstep vibration signals is analyzed by combining the knowledge of geophysics and microseismology.According to the characteristics of footsignal propagation,this paper designs and develops the high precision acquisition instrument for collecting footstep vibration signal,which is a project of national key research and development plan three-CO2 injection and storage state geological and geophysical monitoring technology and equipment(2018YFB0605503)project.It can not only be used for geophysical monitoring,but also for footstep vibration signal.In this paper,MEMS accelerometer with high sensitivity,good anti-noise performance,wide frequency band,small volume and linear phase is selected as the front end of the acquisition instrument.The MEMS sensor and its peripheral circuit are packaged by stainless steel waterproof casing and connected to the collection station through the cable.The collection station mainly carries out the design of signal conditioning,power supply,AD conversion,GPS synchronous acquisition,communication and storage circuit.The conditioning circuit mainly includes the design of filtering and amplifying.The signal from the acquisition instrument can be stored locally via an SD card.In order to monitor the vibration signal in real time,a fast,concise and efficient data real-time acquisition software is developed,which written in mature Visual C++and compiled with an efficient MS VC compiler.The UDP protocol based on high speed and low latency was used to receiving signal from collection station.The user interface is simple and intuitive,and the data can be stored and visualized in real time.Finally,the footstep signal acquisition system is used to design and complete the footstep signal acquisition experiment with 30 testers in the room,and the characteristics of some footstep signals are analyzed.(2)Aiming at the instrument and environmental background noise in the acquisition process,a noise reduction algorithm based on variational mode decomposition and wavelet energy entropy is proposed.The footstep vibration signal is decomposed into multiple intrinsic mode functions by the VMD,and the intrinsic mode functions are decomposed by wavelet transform in multi-scale.The obtained detail coefficients are divided into several sub-intervals to calculate the wavelet energy of each sub-interval.The average value selected from wavelet coefficients of the largest subinterval in wavelet energy entropy,is used as the noise variance of the intrinsic mode function in this scale,then substituted into the threshold formula,and the corresponding threshold is calculated.The threshold function improved by the wavelet energy entropy is used to reduce the noise of the detail coefficients of each intrinsic mode function.After that,the footstep vibration signal was reconstructed by the new intrinsic mode functions.In the process of noise reduction,the selection of VMD parameters is also studied experimentally by calculating the signal-to-noise ratio and program running time.The method of noise reduction in this paper has achieved the highest signal-to-noise ratio in different noise levels and different noise reduction methods by the experiment research.It qualitatively and quantitatively shows the effectiveness of this method and realizes the suppression of footstep vibration signal acquisition instrument and environmental background noise,and maintains the waveform characteristics of footstep vibration signal.(3)After denoising the footstep vibration signal,this paper comprehensively analyzes and extracts the characteristics of the footstep vibration signal from the time,frequency and time-frequency domain.Statistical features are extracted in time domain and frequency domain.At the same time,using the method of endpoint detection,the duration of one-step vibration signal and the interval time of two-step continuous step vibration signal are extracted as time-domain features.The time-frequency domain is analyzed by the simulated and actual collected footstep vibration signals combined with some commonly used time-frequency methods,including Short-time Fourier transform,Gabor transform,Wavelet transform,S transform,Wigner-Ville distribution and so on.Finally,the best S transform is selected,and the distribution of the amplitude of the sample at different frequencies and frequencies amplitude at different sampling points of the footstep vibration signal is analyzed.To solve the problem of high dimension of S-transform time-frequency complex matrix,singular value decomposition method is used to extract the time-frequency characteristics of footstep signal and reduce the data dimension.The footstep signal is divided into different lengths,and then the influence of data length on classification and recognition is studied.Finally,the feature vectors of foot vibration signals with different length are established according to the features extracted in time,frequency and time frequency domain.(4)This paper proposes a feature selection method based on membrane computation and particle swarm optimization combined with SVM.At the same time of optimizing feature subsets,the classifier parameters are also be optimized.The particle swarm algorithm has obvious points that it is easy to prematurely.When the initial velocity of the particle is set incorrectly,it tends to fall into the local optimum.In order to solve this problem,this paper uses the membrane calculation method to optimize the particle swarm algorithm.The distributed and parallel processing characteristics of membrane calculation can better balance the local and global search accuracy of the particles,and also maintain the diversity of particle swarm population.Binary particles are constructed by the penalty parameter and the kernel parameter of classifier,and the number of feature subsets,then converted to the decimal number recognized by the SVM classifier.The fitness function is established by using the classification accuracy and the number of feature subsets.The optimal feature subset and the corresponding classification accuracy are finally output through the optimization iteration of particles.During classification and identification processing,the least square support vector machine and the binary tree structure are used to establish the training model which also be used for testing.After testing through three methods including our method and BPSO-LSSVM,LSSVM,BPNN method,regardless of convergence speed,recognition accuracy and recognition time,the algorithm proposed in this paper shows good results,and effectively identifies different shoe types and different individuals.