A Wearable Acceleration Sensor System for Gait Recognition

This paper addresses the problem of human recognition based on their gait acceleration signal characteristics produced by walking. A portable microprocessor-based data collection device was designed to measure the three-dimensional gait acceleration signals during human walking. The system consists of a tri-axial accelerometer, a MCU, 32 M bytes of RAM, and a data transfer module for data transfer. The device was fixed to the user's waist and three-dimensional acceleration signals were recorded at a sampling rate of 250 Hz. After completing the recording, data stored in the RAM were transferred to a personal computer for wavelet denoising, gait cycles dividing, and gait pattern extracting. Through the analysis in time domain and frequency domain, and using dynamic time warping to deal with the problems result from naturally occurring changes in walking speed, 1-nearest neighbor is used for individual identification. Experiments were performed on 21 subjects walking on their normal speed. The methods of gait analysis in time domain and frequency domain are applying, the equal error rate of 5.6% and 21.1% are achieved respectively. Our preliminary results indicate that it is possible to recognize users based on their gait acceleration.