Gait Feature Authentication Based On Acceleration Sensor

Gait that has uniqueness and stability is one of biometric, and not easy to be forged.Gait feature authentication based on acceleration sensor is an emerging method ofauthentication. As the first step for protecting information security of portable electronicproduct, it does not need the user’s active cooperation, and certification can be finished innatural walking state. Currently gait authentication technology is still in the research stage,mature product has not appeared at home and abroad.In order to improve the efficiency of data collection, a wireless data transmission isused. The design of a wireless gait acceleration data acquisition device takes MPU-6050six-axis motion processing components and JF24D-MCU wireless communication moduleas the core. The system power with battery-powered mode is3.3V, which is compact andeasy to collect. The waist three-axis acceleration data and three-axis gyro data of19subjects are collected during nature walk in the experiment, the sampling rate is166Hz.Acceleration of gravity direction is mainly analyzed.To achieve good recognition results, data preprocessing is needed. This design usesthe low-pass FIR filter to do software filtering after the hardware filtering, to ensure aclean gait signal. Gait cycle is the smallest unit of gait analysis and cycle division is themost important step, which directly affects the effect of the authentication algorithm. Thisdesign uses manual and automatic cycle division methods. The acceleration signal divisionpoints in forward direction are used to automatically divide the acceleration signal in thegravity direction. The autocorrelation coefficient method determines the number of the firststep points, this number is used to automatically classify the cycle. Since during thehuman’s practical walking, number of sampling points from each of the gait cycle is notconstant, so linear interpolation is used to neat the sample for unified template, and theposition of the extremum points from acceleration signal samples is closer. Using the firstderivative of Gaussian function as the wavelet basis function, wavelet transform’s passingzero method can extract feature points. This paper proposes a new dynamic time warping algorithm which makes the timefeature and amplitude feature as a two-dimensional sequence, using pairs of peaks andvalleys to search the minimum cumulative distance path method, the distortion distance,the points of matching, the logarithmic of not matching and the distance of not matchingpairs of peaks and valleys are obtained. The sum of the average distortion distancemultiplied by the coefficient1and the average distance of not matching pairs multiplied bythe coefficient2is total distortion distance of two samples. In a small range of samples, thealgorithm has been verified. The maximum distance inside classes is less than theminimum distance between classes, authentication can be achieved through thresholddetermination.