Body finger print creation using posture, motion and gait recognition for user identification

Abstract

Biometric feature authentications are well known security mechanisms which cannot be steal or hard to imitate. Though many security systems are password protected still can be hacked without user awareness. All existing biometric authentication features require either user’s to carry authentication tokens, remember passwords or physical contact between device and user. Successful user identification mechanism implemented using gait and anthropometric data will help to strengthen the existing biometric authentication systems by combining all together. Providing both behavioral and physiological characteristics allows users to walk freely in restricted areas while authentication is done automatically. The proposed approach is to use the Microsoft Kinect sensor which has the capability of generating the marker less skeleton for walking human and then to apply machine learning techniques to uniquely identify each individuals. Hidden markov models and multiclass support vector machines have been proposed. A prototype was implemented to carry out the model creations and sample validations. Finally experiments have been conducted to evaluate the proposed method using prepared data set which is created in aid of prototype. Dynamic and distance related features have been tested for sample set of 11 humans. There distance related features performed well over the angle related dynamic features while distance related features received the accuracy rate of 63.63% and angle based features received 36.36%. Total combination of both feature sets received the accuracy rate of 72.72%.Since HMM models have performed well over multiclass SVMs .SVMs have failed to retrieve a satisfactory accuracy rate and received only low percentage which is low as 9%.