Research Of Authentication Method Based On Gait Acceleration Signal

Biometric systems, such as fingerprint, iris identification, face recognition and voice verification system have been wildly used in many areas as they can offer more reliable and efficient ways of individual identification than traditional methods like using PIN code. However, these commonly used methods require users' actions and can only provide interval identification in specific stages such as power-on or unlock. As a physiological characteristic, gait information can be extracted automatically during human walking, thus it is possible to give continuous protection to portable electronics.A gait acceleration measurement device is designed in this paper. The device is equipped with a MEMS accelerometer. It uses two high-speed Ferroelectric RAMs and a NAND Flash memory chip to constitute two-level memory architecture, and two microcontrollers are employed for parallel control of data collection and storage. This acceleration measurement device has a pre-eminence of small size, light weight and easy to operate. With this device, a gait acceleration database is created that contains 24 volunteers' acceleration data acquiring from back of waist and 22 volunteers' data from their right ankle. Researches of waist-based and ankle-based authentication methods are carried out using this database.In waist-based authentication method, gait features are extracted from the original acceleration signals in vertical direction using zero-crossings of wavelet transform algorithm based on Gaussian wavelet, and represented as seven-feature-tuple sequences. Dynamic time wrapping algorithm is employed to match the feature sequences from different samples and calculate the distortions. Based on these distortions, a multi-criterion model is designed for authentication, and the equal error rate of this method is 5.0%. In ankle-based authentication, a simplified method called feature point method is developed. The feature points are extreme points of acceleration signals in backward-forward and vertical directions. Zero-crossings of wavelet transform algorithm is used to extract feature points. Dynamic time warping (DTW) algorithm is employed to match the feature points and calculate the distortions, and a multi-criterion model is designed to identify individuals. The equal error rate of authentication has achieved 3.27% with this method.Since the feature points are corresponding with the gesture like heel contact, foot flat, initial push-off, etc. Statistic research shows that different person have different force character during these gestures. The features extracted in this paper capture the essence of the individual gait characteristics. Dynamic time wrapping algorithm is used to realized a non-linear matching of the features from different gait samples, which eliminated the impacted of speed of non-linear changing. The experimental results show that the equal error rates of these two gait authentication methods are better than that of the previous literatures reported.