A speed invariant human identification system using gait biometrics

摘要

Can gait biometrics be used for identification of a person? We feel that each individual has an intrinsic gait behaviour, irrespective of the individual's gait speed. The challenge is to extract this gait behaviour from the gait biometrics. In this paper, we used a computer vision-based technique for gait identification. The silhouette treadmill gait database obtained from OU-ISIR, Japan has been used in this gait research work. We have used 22 subjects walking at different speeds varying from 2 km/hr to 6 km/hr with speed variation of 1 km/hr. The gait energy image (GEI) has been computed from this gait data. The width of GEI, along the horizontal axis, has been used as the feature vector for training and testing. These features show speed invariance but is intrinsic and unique to the subject. The feature captures the intrinsic hand movement, head node and leg oscillations of the subjects. A probabilistic model based on Baye's conditional probability rule and connectionist model based on multilayer perceptron neural network have been used for classification. This technique provides a promising result of identifying a subject invariant of the gait speed.