Gait Authentication under non-standard environments
Humans have for ages used characteristics on a human to identify and authenticate
people, either by face, voice, fingerprints, etc. In recent time human recognition
has become an important task in a variety of applications, such as access control and
surveillance. Authentication can happen in many ways, but all authentication factors
can be categorized into one of three classes:
In this paper we will look at something you are, which consequently utilizes biometric
features of a person. Biometric features can be divided further into two main categories:
When using physiological biometrics the user must usually interact with a system, for
instance scanning his fingerprint on a fingerprint reader. With behavioral biometrics the
data that are collected can normally be recorded when the user performs his natural
duties, for instance talking in a phone. Even though behavioral biometrics required less
user interaction, it is physiological biometrics that is used most. The reason for this is
that physiological biometrics perform in general better when a system is authenticating/
identifying a user. This fact leads to the growing interest to improve behavioral biometrics.
This project will look at gait as a biometric feature. Gait recognition has become an area which has gained a lot of interest the last decade. An important reason why gait has become attractive is that is non-intrusive, can be measured without subject contact or knowledge and it can not easily be obscured [5]. The most research on gait recognition the last decade has been video-based, where the purpose has been for surveillance, for instance recognizing a criminal from a security camera video. There has also been some research on sensors installed in the floor which can be used in an access control system. When using video-based recognition however, there are a lot of variables interfering which lower the performance, such as light and other objects. In 2005 another identification method which utilize how people walk by looking at acceleration
Gait recognition under non-standard environments from sensors attached to the belt was presented. This method uses a device called an accelerometer which measures acceleration in three directions (horizontal, vertical and lateral), and such a device will be used in this project. Gait recognizing using accelerometer can be used to authenticate and protect mobile phones and other portable electronic devices, where the sensors could be integrated into the hardware. As with video-based recognition, there are some circumstances which lower the performance of accelerator based authentication. A common problem with this technology is the fact that one get noise that affects your signal, another problems are that your gait would be altered if you are injured, drunk and so on.
In this project we will investigate how gait recognition works under non-standard environments. We will try to, by isolating variables, find out how different variables impact the recognition rate. Furthermore will there be an analysis to see if the data collected under different circumstances can be adapted in order to be better comparable with each other.
Gait recognition is normally achieved after walking back and forth on a solid surface and in a straight line. In a real life scenario this is unfortunately not the case, people walk on different surfaces, walk up and down stairs, walk with different speed, use different clothing and shoes, wear or carry backpacks/briefcases etc. These real-life circumstances introduces some challenges when trying to authenticate people by their gait. In that connection would it be interesting to see how gait recognition works under non-standard environments. It would be desirable that by e.g. knowing what kind of surface a person walks on or if a person is walking faster/slower than normal, the data collected by the accelerometer could be adjusted accordingly and compared with the baseline. The baseline which are being compared to is recorded during an enrollment phase under normal circumstances. In the recognition process the system must try to recognize the current situation and transform the data in a standard manor before comparing with the baseline.
In todays world portable electronic devices such as mobile phones and PDAs have become a natural and important tool. This technology has exploded during the last years, you do not have to go many years back before for instance the mobile phone was just a communication device. Today however, mobile phones are used in applications like m-banking and m-government. With this in mind it is not hard to realize the consequences if your mobile phone is stolen or lost, the financial and personal data would now be accessible by the thief. Currently the only protection mechanism which resides on such devices is usually a PIN code and perhaps a password, the need for a better security is obvious. Features like fingerprints and voice has been proposed with various results, but these features are either obtrusive, require user’s attention or merely did not perform satisfactory. The use of voice based recognition did actually perform very well Gait recognition under non-standard environments under low-noise circumstances, but had more problems with higher background noise. As a result of this gait has come up as an additional way to secure your phone. With gait the system has the possibility to perform continuously authentication. If gait recognition shall become an integrated part of securing electronic devices however, must it be able to perform adequate under different conditions.
In order to solve the problem outlined there are first some general issues that must be considered:
Those questions I will look deeper into during this project are:
This project will come up with results how some different circumstances affect the ability to recognize people. If the circumstances do not significantly impact the recognizing process or if it would be possible to do some special processing according to what the circumstance are, there should be no problems using a training set from indoor walking for continuous authentication. This means that after the enrollment phase, the authentication could more easily be moved to another place without having to train the system Gait recognition under non-standard environments again to accommodate for different circumstances. This would be a huge step in order to secure mobile devices with gait analysis.
