Gait Recognition Based on Deep Learning Using Accelerometer and Gyroscope in Smartphones

Abstract

Gait Recognition is difficult when compared to other biometrics and has an unobtrusive advantage, but it has little interaction among the users. The gait dynamics are captured using accelerometer and gyroscope, which are the initial sensor used in the smartphones. The gait data is inexpensive and convenient to collect using the inertial sensor integrated in the smartphones which an average person can commonly use. The gait recognition utilizing the smartphones in the wild is proposed in this paper. The traditional method for gait recognition requires a person walking on a specified road, walking speed etc. The inertial data is collected for gait recognition under a free situation without knowing the knowledge of data collection in a user walk. The deep learning techniques are utilized to obtain authentication performance and person identification to learn the gait based on a biometric model based on the person walking data. The robust gait feature representation is obtained using the proposed hybrid deep learning technique. The space and time domain is successively abstracted by the convolution neural network.