Typescript (photocopy). / Digitized by Kansas Correctional Industries

This thesis presents the work carried out on static and dynamic analysis of near infra-red dorsal hand vein images for biometric applications. It focuses on the investigation of the classification of old and young groups of people, and analyses vital signs underlying dorsal hand vein images. All research is based on the database used in this work. Focussing on the aged group of dorsal hand vein images, a database including static images and video images was first created. The static database contained 1000 images from 50 individuals and the dynamic database included 40 videos from 20 old participants. For static analysis, dorsal hand vein images were pre-processed by geometry correction, regions of interest extraction (ROI), grey level normalisation, noise reduction and image enhancement. Then, skin and vein areas from the dorsal hand were segmented using maximum curvature based algorithms. Due to varying haemoglobin and water levels in the vein and skin, intensity based parameters were investigated and extracted as features for the classification of old and young groups. Two classifiers, linear discriminant analysis (LDA) and k-nearest neighbours (KNN) were adopted for comparative discussions. The experimental results turned out to be satisfactory and the two groups were well classified, using statistical intensity based features. For dynamic analysis, mean grey levels were extracted from the ROI of each frame of the dorsal hand vein image videos. Then, all parameters were connected to form a biometric signal. The signal was analysed in a spectrum to detect a major peak for liveness. A fake dorsal hand video was introduced for comparative studies. Experimental results showed that a respiratory like signal was detected as the main peak in the spectrum, verifying the vital signs of dynamic dorsal hand vein images and proposing a new method of liveness detection in biometric applications.

Individuals who have upper limb movement problems include people with cere­bral palsy (CP) and stroke victim . Both thes conditions lead to difficulties in daily activities such as reaching, gra ping etc. Loss of motor function can be improved or recovered via relevant target specific movement rehabilitation. This r search aims to d velop and evaluate an electromyography (EMG) biofeedback based virtual reality (VR) system targeted towards CP and stroke patients that could provide an engaging and effective way to practice and improve hand and arm motion functions. VR, which could provide a repetitive multimodal task­orient d r habilitation environment for patients to undertake elf-training in safety, is con id red to be a suitabl tool for medical health rehabilitation. Using EMG biofeedback in rehabilitation could provide pati nts with opportunities to improve their ability by ass ssing th ir muscle activity r spons and learning self-control of mov m nt during specific training tasks. Th syst m development incorporat s concepts f om motor learning, visual art and programming, and provid s visual and audio feedback that guides the us r' move­m nt to becom smoother and more fficient. Using compo ite multiple feedback in upp r limb r habilitation enables th individual with upp r limb probl m to per­ceiv similar r al-world p rformanc in th virtual world. Howev r, it i r cognised that such fe dback might cause information overload, 1 ading to the individual feel­ing confused and distract d during training. The necessity of involving healthy subject in f asibility tudies is also recognised. Therefore, thi study focuses on healthy subjects to investigate the effectiveness of separate function-specific fe d­back training exercises in order to overcome the information overload problem. Two s ts of xp riments based on 2D display and 3D display were carried out in thi research. A total of sixty healthy subjects participated in the experiments, thirty for the 2D display tests and thirty for the 3D display tests. The developed EMG biof edback based VR system provided the participants with position, orientation, and muscle activity response information. This allow d combined task-oriented multiple feedback which aimed to improve the performance in the virtual environment. In order to preempt and deal with the information overload problem, three prior individual fe dback pre-training game-like tasks were designed based on the distinctive functions of the individual VR interactive inter­faces (magnetic track r and EMG system). As in video games, where preparative modes are provided, and the player can learn the key functions and the game sce­nario through prior practice, so as to improve their performance in the full game. This allows the multimodal and multiform fe dback training exercises are broken down into individual function-specific feedback exercises. The training tasks were designed in the form of gam -like tasks, which could offer better motivation and allow easier hierarchical delivery. Candle lighting, ball balancing and cup grasping game-like tasks, which are position movement, orientation movement and muscle contraction pre-training tasks respectively, were designed based on th distinctive function of magnetic tracking and EMG systems. The balloon shooting main game, which is re-engineered ba ed on a tutorial provided by th XNA game studio, was developed for th multiform feedback training task. The d veloped game-like VR rehabilitation tasks could make th information from the biological level visible to participant with multimodal and multiform feedback. In both 2D and 3D display tests, th r are two experimental groups with pre-training prior to th main game and a control group with th main task only. The empirical results of both 2D and 3D display t t indicate that the developed EM G biofe d back based VR system and tasks are useful to improv the ability of ome of th upp r limb activitie , and th prior individual feedback pre-training tasks are h lpful for b tt r performance in the main multiform fe dback task . This tudy show d that the information ov rload occurring in th integrat d VR syst m with multiple forms of£ edback can b ov rcom with prop rly design pre -training tasks.

Thesis (Ph. D.)--University of Texas at Austin, 2005. / Vita. Includes bibliographical references.

University of Macau / Faculty of Science and Technology / Department of Electromechanical Engineering

Not available

No description available.

No description available.

No description available.

No description available.