1. UKM Learning and Research Repository
2. Master Thesis / Tesis Master
3. Faculty of Information Science and Technology / Fakulti Teknologi dan Sains Maklumat

Title:	Real-time eye gaze tracking on poor quality video using template matching and hough transform
Authors:	Mohammad Sadegh Sarrafi (P50168)
Supervisor:	Md. Jan Nordin, Prof. Dr.
Keywords:	Eye gaze Poor quality video Template matching Hough transform Human-computer interaction
Issue Date:	28-Mar-2011
Description:	Eye gaze tracking is a very high potential field in human and machine interaction. Eye detection and gaze tracking have been very intensive research area for decades. Eye gaze is known as the direct of a person’s sight and it is related to the person focus of attention before its action. Therefore it can be a promising technique for human and computer interaction. There are many different kind of application have been developed in variety of usage such as eye disease diagnosis, human behavior study and human computer interaction. These systems have been developed in various numbers of techniques such as computer base vision, electronic or magnetic methods. However this technology is not widely used to develop applications due to difficulties of usage for users, such as wearing additional equipments or using high quality and expensive cameras and other devices. However this research is going to find a method which is simpler and easier to develop and utilize. Also is trying to reduce the development cost and find a method to be more affordable and more comfortable for users to use. This research tries to find a solution for this technology to make it available for home users on personal computers by using an ordinary webcam to track the eye. The main concern of this research is focused on improvement of stability and accuracy of eye detection and gaze detection in real time on poor quality videos and images. This research has an improvement on eye detection from a whole image of face using template matching. In this technique the dimension of eye template is reduced to iris area. In this way the difficulty of matching a variety of eyes’ shapes and also the number of eye samples are reduced. This improvement is examined by an experimental testing which results to 74.28% of accuracy in this technique. And also this work has an enhancement in pupil detection using Hough Transform algorithm to compare with previous works using the same techniques which results to 92.98%. The Hough algorithm has been used in most of technique to detect the shape of pupil however this work has added some filters to this technique to have more accurate and stable detection. The data have been used in this research are from different datasets. The most popular dataset in eye tracking researches is from CASIA database which have been used in this research as well, but since the CASIA’s datasets are from very high quality and closed frame of eye therefore they can’t be suitable for this work, hence other datasets which are containing less quality images in variety of size and gesture also have been used in this research. The achieved results are experimented from very poor images from (The Database of Faces) and (Faces in the Wild) which are available online for academic usage. The main platform is VC++ and OpenCV library have used in this work to build a prototype in purpose of experimental testing.,Master/Sarjana
Pages:	104
Call Number:	QA76.9.H85.S246 2011 3
Publisher:	UKM, Bangi
Appears in Collections:	Faculty of Information Science and Technology / Fakulti Teknologi dan Sains Maklumat

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.