AbstractA virtual world is an interactive 3D virtual environment that visually resembles complex physical spaces, and provides an online community through which the users can connect, shop, work, learn, establish emotional relations, and explore different virtual environments. The use of virtual worlds is becoming popular in many fields such as education, economy,space, and games. With the widespread use of virtual worlds, establishing the security of these systems becomes more important. To this date, there is no mechanism to identify users of virtual worlds based on their interactions with the virtual world. Current virtual worlds use knowledge-based authentication mechanisms such as passwords to authenticate users. However they are not capable of distinguishing between genuine users and imposters who possess the knowledge needed to gain access to the virtual world.
The aim of the research reported in this thesis is to develop a behavioural biometric system to identify the users of a virtual world based on their behaviour inside these environments. In this thesis, three unique virtual worlds are designed and implemented with different 3D environments and avatars simulating the different environments of virtual worlds. Two experiments are conducted to collect data from user interactions with the virtual worlds. In the first experiment 53 users participated and in the second experiment, a year later, 66 different users participated in the experiment.
This research also studies the parameters of user behaviour inside virtual worlds and presents novel feature extraction methods to extract four main biometric features from the collected data, namely: action, time, speed, and entropy biometric features. A sample classification methodology is formulated. Using distance measure algorithms and based on the collected data, users are identified inside the virtual worlds. Also in this thesis the application of biometric fusion in enhancing the performance of the behavioural biometric system is studied.
The achieved average equal error rates in this research were between 26-33% depending on the virtual world environment and movement freedom inside virtual worlds. It has been found that avatar actions inside virtual worlds carry more identifying attributes than parameters such as the avatar position inside the virtual world. Also it has been found that virtual worlds with very open environments with respect to avatar movement showed higher EERs when using the biometric system implemented in this research.
|Date of Award||Jul 2012|
|Supervisor||Nick Savage (Supervisor), David Ndzi (Supervisor) & Rinat Khusainov (Supervisor)|