Gait Recognition Based On Mobile Acceleration Sensor

Gait recognition is a new type of biometric recognition technology,which attracts more and more attention.With the rapid development of sensor technology,the research direction of gait recognition technology is changing from computer vision to mobile terminal.In order to improve the security of mobile devices,biological recognition technologies such as fingerprint recognition and face recognition have been proposed.However,the existing methods need to perform the corresponding posture recognition.Gait signals captured by mobile acceleration sensors can be effectively implicitly authenticated.However,when the gait signal is collected,due to factors such as shaking and bumping,the mobile phone tends to be unstable in the direction.Therefore,a new scheme is proposed to solve the instability problem of sensors.At the same time,in view of the low detection accuracy of the traditional gait cycle detection algorithm,An autocorrelation peak detection algorithm is proposed to find the gait cycle accurately.In addition,the traditional gait recognition is mostly based on template matching.The recognition accuracy of this scheme is not high.Therefore,this paper proposes a new gait recognition method,which uses statistical analysis and supervised learning to recognize.The scheme is superior to the traditional scheme,and achieves 4.7% equal error probability and 90.45% accuracy in authentication and recognition,respectively.The contents of this paper are as follows:1)The original gait data are preprocessed,including coordinate system correction,gravity acceleration component filtering,linear interpolation,denoising and so on,so as to obtain a relatively pure gait signal.2)Gait cycle was extracted for pre-processed signal,this paper proposed a new gait cycle detection algorithm-autocorrelation and peak detection algorithm,which can accurately find the gait cycle.3)The gait signal is divided into gait patterns one by one,4 gait cycles are one gait mode,and there is a 50% overlap between gait patterns.According to research by Derawi,Gafurov et al.,it was found that the amplitude of the signal can improve the recognition rate.Therefore,two new dimensions are created for the gait signal,which are the combination of the X and Y dimensions and the combination of the X,Y,and Z dimensions.4)Extract the eigenvectors of each gait pattern in each dimension,Learning from the feature extraction method of speech recognition,The Mel frequency cepstral coefficient(MFCC)and the Bark frequency cepstral coefficient(BFCC)are extracted.The feature vector extracted in each dimension,take three dimensions as a combination to form a feature vector of the entire gait recognition system.For subsequent classification and recognition,the extracted feature vectors are first subjected to PCA dimensionality reduction,then renormalization processing.5)To classify and recognize feature vectors,BP neural network and support vector machine classifier are established in this paper.6)In order to verify the feasibility of the proposed scheme,experiments on recognition performance and authentication performance are carried out.The experimental results show that the proposed scheme can obtain higher recognition rate and lower equal error rate compared with the existing schemes.and it has wide application value.