Patrick Bours

Unobtrusive User-Authentication on Mobile Phones Using Biometric Gait Recognition

The need for more security on mobile devices is increasing with new functionalities and features made available. To improve the device security we propose gait recognition as a protection mechanism. Unlike previous work on gait recognition, which was based on the use of video sources, floor sensors or dedicated high-grade accelerometers, this paper reports the performance when the data is collected with a commercially available mobile device containing low-grade accelerometers. To be more specific, the used mobile device is the Google G1 phone containing the AK8976A embedded accelerometer sensor. The mobile device was placed at the hip on each volunteer to collect gait data. Preproccesing, cycle detection and recognition-analysis were applied to the acceleration signal. The performance of the system was evaluated having 51 volunteers and resulted in an equal error rate (EER) of 20%.

Improved Cycle Detection for Accelerometer Based Gait Authentication

Over the last years, there has been an increasing research interest in the application of accelerometry data for many kinds of automated gait analysis algorithms. The need for more security on mobile devices is increasing with new functionalities and features made available. To improve the device security we propose an improved biometric gait recognition approach with a stable cycle detection mechanism and comparison algorithm. Unlike previous work on wearable gait recognition, which was based from simple average cycling methods to more complicated methods, this paper reports new techniques for which can improve the performance, by using simple approaches. Preprocessing, cycle detection and recognition-analysis were applied to the acceleration signal. The performance of the system was evaluated having 60 volunteers and 12 sessions each volunteer and resulted in an equal error rate (EER) of 5.7%.

Spoof Attacks on Gait Authentication System

Research in biometric gait recognition has increased. Earlier gait recognition works reported promising results, usually with a small sample size. Recent studies with a larger sample size confirm gait potential as a biometric from which individuals can be identified. Despite much research being carried out in gait recognition, the topic of vulnerability of gait to attacks has not received enough attention. In this paper, an analysis of minimal-effort impersonation attack and the closest person attack on gait biometrics are presented. Unlike most previous gait recognition approaches, where gait is captured using a (video) camera from a distance, in our approach, gait is collected by an accelerometer sensor attached to the hip of subjects. Hip acceleration in three orthogonal directions (up-down, forward-backward, and sideways) is utilized for recognition. We have collected 760 gait sequences from 100 subjects. The experiments consisted of two parts. In the first part, subjects walked in their normal walking style, and using the averaged cycle method, an EER of about 13% was obtained. In the second part, subjects were trying to walk as someone else. Analysis based on FAR errors indicates that a minimal-effort impersonation attack on gait biometric does not necessarily improve the chances of an impostor being accepted. However, attackers with knowledge of their closest person in the database can be a serious threat to the authentication system.

Gait Authentication and Identification Using Wearable Accelerometer Sensor

This paper describes gait recognition using a body worn sensor. An accelerometer sensor (placed in the trousers pocket) is used for collecting gait features. From the acceleration signal of the person, cycles have been detected and analysed for recognition. We have applied four different methods (absolute distance, correlation, histogram, and higher order moments) to evaluate performance of the system both in authentication and identification modes. Our data set consists of 300 gait sequences collected from 50 subjects. Absolute distance metric has shown the best performance in terms of EER, which is equal to 7.3% (recognition rate is 86.3%). Furthermore, we have also analysed recognition performance when subjects were carrying a backpack.

Continuous keystroke dynamics: A different perspective towards biometric evaluation

In this paper we will describe a way to evaluate a biometric continuous keystroke dynamics system. Such a system will continuously monitor the typing behaviour of a user and will determine if the current user is still the genuine one or not, so that the system can be locked if a different user is detected. The main focus of this paper will be the way to evaluate the performance of such a biometric authentication system. The purpose of a performance evaluation for a static and for a continuous biometric authentication system differ greatly. For a static biometric system it is important to know how often a wrong decision is made. On the other hand, the purpose of a performance evaluation for a continuous biometric authentication system is not to see if an impostor is detected, but how fast he is detected. The performance of a continuous keystroke dynamic system will be tested based on this new evaluation method.

Gait and activity recognition using commercial phones

This paper presents the results of applying gait and activity recognition on a commercially available mobile smartphone, where both data collection and real-time analysis was done on the phone. The collected data was also transferred to a computer for further analysis and comparison of various distance metrics and machine learning techniques. In our experiment 5 users created each 3 templates on the phone, where the templates were related to different walking speeds. The system was tested for correct identification of the user or the walking activity with 20 new users and with the 5 enrolled users. The activities are recognised correctly with an accuracy of over 99%. For gait recognition the phone learned the individual features of the 5 enrolled participants at the various walk speeds, enabling the phone to afterwards identify the current user. The new Cross Dynamic Time Warping (DTW) Metric gives the best performance for gait recognition where users are identified correctly in 89.3% of the cases and the false positive probability is as low as 1.4%.

Improved Gait Recognition Performance Using Cycle Matching

This paper presents a user authentication method based on gait (walking style). Human gait (in terms of acceleration signal) is collected using a wearable accelerometer sensor attached to the ankle of the person. Ankle accelerations from three directions (up-down, forward-backward and sideways) are utilized for identity verification purposes. Applying cycle matching method on experimental data from 30 subjects, we obtained an encouraging EER (Equal Error Rate) of 1.6% using sideway acceleration signals. In addition, our analysis indicate that performance is better with light shoes than with heavy shoes; and sideways motion appear to provide higher discrimination power compared to the up-down and forward-backward motions. Application area for such gait-based user authentication approach can be in improving security and usability of user authentication in emerging applications such as pervasive environment.

Scenario test of accelerometer-based biometric gait recognition

The goal of our research is to develop methods for accelerometer-based gait recognition, which are robust, stable and fast enough to be used for authentication on mobile devices. To show how far we are in reaching this goal we developed a new cycle extraction method, implemented an application for android phones and conducted a scenario test. We evaluated two different methods, which apply the same cycle extraction technique but use different comparison methods. 48 subjects took part in the scenario test. After enrolment they were walking for about 15 minutes on a predefined route. To get a realistic scenario this route included climbing of stairs, opening doors, walking around corners etc. About every 30 seconds the subject stopped and the authentication was started. This paper introduces the new cycle extraction method and shows the Detection Error Trade-Off-curves, error rates separated by route-section and subject as well as the computation times for enrolment and authentication on a Motorola milestone phone.

Eigensteps: A giant leap for gait recognition

In this paper we will show that using Principle Component Analysis (PCA) on accelerometer based gait data will give a large improvement on the performance. On a dataset of 720 gait samples (60 volunteers and 12 gait samples per volunteer) we achieved an EER of 1.6% while the best result so far, using the Average Cycle Method (ACM), gave a result of nearly 6%. This tremendous increase makes gait recognition a viable method in commercial applications in the near future.

Renewable Minutiae Templates with Tunable Size and Security

A renewable fingerprint minutiae template generation scheme is proposed to utilize random projection for template diversification in a security enhanced way. The scheme first achieves absolute pre-alignment over local minutiae quadruplets in the original template and results in a fix-length feature vector; and then encrypts the feature vector by projecting it to multiple random matrices and quantizing the projected result; and finally post-process the resultant binary vector in a size and security tunable way to obtain the final protected minutia vicinity. Experiments on the fingerprint database FVC2002DB2_A demonstrate the desirable biometric performance of the proposed scheme.