Authentication based on a changeable biometric using gesture recognition with the Kinect™

Abstract

Biometric systems either use physiological or behavioural characteristics to identify an individual. However, if a biometric is compromised it could be difficult or impossible to change it. This paper proposes a biometric authentication system based on gesture recognition, where gestures can be easily changed by the user. The system uses a Kinect™ device to capture and extract features, as it provides 20 skeleton tracking points: we use just six of these in our system. The Dynamic Time Warping (DTW) algorithm is used to find an optimal alignment between gestures which are time-bound sequences. We tested the system on a sample of 38 volunteers. Ten volunteers provided reference gestures of their own design and 28 volunteers attempted to attack these reference gestures by both guessing and copying. Guessing the gesture was unsuccessful in all cases, but when the attacker had previously seen a video of the reference gesture the experiment gave us an estimation of the True Positive Rate (TPR) of 0.93, False Positive Rate (FPR) of 0.017 and Equal Error Rate (EER) of 0.028.