WWW.DISSERTATION.XLIBX.INFO
FREE ELECTRONIC LIBRARY - Dissertations, online materials
 
<< HOME
CONTACTS



Pages:     | 1 |   ...   | 17 | 18 || 20 | 21 |

«BY GEORGIOS DIAMANTOPOULOS A THESIS SUBMITTED TO THE UNIVERSITY OF BIRMINGHAM FOR THE DEGREE OF DOCTOR OF PHILOSOPHY DEPARTMENT OF ELECTRONIC, ...»

-- [ Page 19 ] --

This paper is concerned with the development of the Robust Eye-Accessing Cues Tracker (REACT) whose main requirements are to maintain a high level of accuracy while tracking nonvisual eye-movements as well as minimize invasiveness and cost. The REACT eye-tracker is head-mounted but very lightweight (approx. 60 grams) and is thus minimally invasive. A headmounted approach was chosen over a remote camera one as to minimize cost and increase the resolution of the captured images and consequently the accuracy viable by the eye-tracker.

Moreover, a remote camera requires that the head is tracked and not only does that increase the computational complexity but it also decreases the ability of the eye-tracker to track extreme eye-movements as discussed earlier. The hardware design is based on the low-cost eye-tracker presented in [1], with the scene camera and associated hardware removed. As such, no more time will be spent discussing the hardware design itself.

233 On the hardware-level, the eye-tracker works by illuminating the eye with infrared light while blocking most of the visible light spectrum with an infrared filter imposed over the camera lens.

This produces the dark-pupil effect which allows for the easy detection of the pupil and, in a sense, offloads some of the processing to the hardware. The captured images are then processed in software and a set of three eye features are detected: the pupil centre and contour, the iris radius and the location of the eye corners. Finally, features are combined in order to calculate the 2D gaze angle and classify each eye-movement to one of eight eye-movement classes extracted from the EAC model [2]: up, up and left, left, down and left, down, down and right, right and up and right.

This paper is organised as follows. Section 2 describes the extraction of the eye features (pupil centre and contour, iris radius, and eye corners) in detail. Section 3 describes the calculation of 2D gaze and its classification to one of eight classes. Section 4 is concerned with the performance evaluation of the eye-tracker and Section 5 describes a pilot study experiment that was designed to serve as a real-world case study. Some concluding remarks are offered in Section 6.

2 Eye feature extraction

Active illumination from one or more infrared sources is a common method used in eye-tracking [7] [12] [15] to produce the dark- or bright-pupil effect. The bright-pupil effect is produced when the light source is coaxial to the camera and the dark-pupil effect otherwise [6]. Either effect is particularly useful in locating the pupil as it is highly-contrasted to the iris which has different reflection properties and thus appears as grey, as shown in Figure 1 (in this particular setup, it is the dark-pupil effect).

–  –  –

{ Using connected components labelling, the binary image can be converted to a higher-level description of the image content. Then, a minimum and maximum blob area is defined ( and respectively) and used to filter blobs resulting from random noise (too small) and blobs that are formed in the darker areas of the image (where the infrared illumination fades rapidly).

–  –  –

Pupil contour refinement After performing an initial estimation of the pupil contour from the thresholded image, it is necessary to further refine it; since global thresholding is used, the results are very much dependent on the threshold value used. Whilst the pupil is always dark in the images, just how dark it is will depend on the camera topology (distance from the eyeball and angle of positioning i.e. exactly how the infrared light falls on the pupil), its pupil reflection properties and their variance between subjects. The aforementioned refinement is done in the REACT eye-tracker through the use of active contours [3], also known as snakes. The greedy snake algorithm implemented in the REACT eye-tracker is derived from [19].

–  –  –

Locating the iris is a problem that has been tackled several times before but usually in remote eye-tracker setups with significantly different image formations (e.g. [14] [17]). Thus a novel approach is proposed here.

Even though it may have been possible to design the eye-tracker without locating the iris boundaries, doing so provides two significant advantages: a) it provides a robust starting point for the challenging task of locating the eye corners which is a very challenging task and b) it allows the eye-tracker to be extended to calculate the 3D gaze provided the camera is fully calibrated; one such approach is found in the remote eye-tracker system developed in [17].

A similar task to the detection of the iris boundaries, is cell segmentation [22]. Cells, much like the iris, are fairly uniform in terms of the pixels’ intensity levels on the inside. In a similar vein, cell image background is also uniform, like the sclera. Thus, the edge is not necessarily defined by the change in grey-level intensity, which is the basis of most edge-detectors, but rather by a change in the uniformity over a range of pixels.

For a population of size, the edge strength at a division, is defined as [22]:

–  –  –

Where ̂ is the standard deviation of the grey-level pixels.





In the original edge strength calculation equation shown above [22], the whole population is considered; however, it was empirically found that with eye image data, edges attributed to eyelashes and eyelids can severely alter the standard deviation of each population and thus make the algorithm fail or return erroneous (in this context) results. Thus, the original formula was modified to work with a fixed window such that for a line of pixels, given a constant window

size, the edge strength at a index is equal to:

–  –  –

The iris is detected on frames where the person is looking approximately straight ahead and several boundary candidates are generated using the above approach. A filtering algorithm

–  –  –

Eye corner detection Locating the eye corners is probably the most significant challenge for the set of input images taken with the REACT eye-tracker. The problem of locating the eye-corners has been tackled before [9] [11] [16] [18] [20] [21] but the systems in question operated, without exception, on a full-face, sometimes colour, image.

Detecting the location of the corners is also a significant task as it provides two static points of

reference. In the current configuration, these two reference points are useful in two ways:

(c) To calculate the principal axis by which to calculate the 2D gaze angle (next section). The camera may be rotated as a result of misplacement by the experimenter or slippage of the frame.

(d) In long sequences where the absolute position of the eyeball centre is bound to change over time (e.g. frame slippage etc.), the eye corners can be used to detect whether a reinitialization of the eye tracker is required.

Additionally, if the eye-tracker was to be configured to detect the 3D gaze (using a fully calibrated camera), the eye corners are essential to disambiguating the 3D vector solution. For more details, the interested reader is referred to [17].

In order to ease the task of finding the eye corners, some detail as well as noise is removed; the input image is scaled down to ¼ of its original size using Gaussian Pyramid Decomposition. Then,

the partial x- and y-derivative of the scaled image are calculated:

| | | | For the y-derivative, non-maxima are suppressed locally using a 1x3 window. Whilst this is an irregular window (usually square windows are used for computer vision operations), it has been

–  –  –

As mentioned earlier, slightly different algorithms are used to detect the inner and outer corner due to the different eye morphological structure evident at this image resolution. Both

algorithms are however based on the same principle:

5. It is assumed that the edges formed between the eyelids and the sclera are within the top local maxima for a restricted window ( ). This assumption was empirically tested.

6. A grouping process begins near the iris boundary previously found and continues outwards, grouping all local maxima that are connected, using an 8-connectivity criterion.

7. The groups are searched for a set of two predefined patterns (shown in Figure 3) and if found, the grouping is terminated at that point. These patterns have been empirically found to occur when the lower eyelid edge is joined with another face line edge and thus the purpose of this step is to separate the two edges.

8. The final corner is selected from the group (outer corner) or pair of groups (inner corner) that demonstrate the maximum derivative energy. The energy of a group of

points is calculated as:

–  –  –

The added steps and differences between the two algorithms are briefly summarized here. In Step 1, the search window includes both the upper and lower eyelid edges for the inner corner

but only the lower eyelid edges for the outer corner. This is done for several reasons:

c) For the inner corner: typically, the lower eyelid edge does not meet the upper eyelid edge. Thus, a distance criterion between the two edges has to be applied to find the corner. Further, often, the lower eyelid edge will be joined to a face line. The pattern detection offered in the main algorithm will in some cases alleviate this problem but only in combination with the distance constraint is the algorithm robust.

239

d) For the outer corner: typically, the camera is rotated around the vertical axis towards the outer corner thus making the eyelid-half on the inner corner side appear longer and the eyelid-half on the outer corner side appear shorter. Thus, the upper eyelid on the outer corner side is sloped several degrees more than the inner corner side. For this reason, grouping local maxima points on the top eyelid results in several disjointed groups. In order to robustly find the outer corner, pattern matching is combined with a refinement based on the partial x- derivative of the image. Since the upper eyelid edge and the lower eyelid edge always meet on the outer corner side, a strong maxima is created in the partial x-derivative image. This maxima is used to refine the corner in the last step and is found by searching a 10x5 window.

FIGURE 3: PREDEFINED SEARCH PATTERNS FOR THE OUTER CORNER SEARCH. THE PATTERNS ARE

REVERSED FOR THE INNER CORNER.

Finally, to increase the robustness of the corner detection results which can be sensitive to noise, corner detection is applied over a group of consecutive frames centered on the target frame and outliers in the resulting eye corners are calculated by the same statistical process used to filter iris boundary candidates. During development, it was found that this algorithm gave superior performance versus no clustering and a weighted means scheme where candidates are inversely weighted according to their distance to the mean and the weight sum is calculated.

3 Calculation and classification of 2D gaze

Calculating a 3D gaze vector would have required a fully calibrated camera (Wang et al., 2000).

Even though there are modern means of camera calibration that greatly simplify the process, it is still too involved to be performed by the user of a system (versus the developer) like the REACT

–  –  –

The eye-tracker requires only one calibration point – that of the subject looking approximately straight ahead. This calibration point can be provided on-line (in real-time while recording and tracking at the same time) or off-line (after the data sample has been recorded) and initializes the tracker by calculating the initial pupil position and contour, the iris radius and the eye corners locations and.

At each point in time, the 2D gaze vector is calculated as follows:

–  –  –

In addition to the evaluation of the feature extraction and 2D gaze calculation algorithms, it was decided to demonstrate the eye-tracker’s applicability to a real-world scenario through a case study. Bearing in mind our motivation through the NLP literature, a pilot study was designed where a subject was interviewed by the experimenter on various topics such as education and hobbies, meanwhile recording the subject’s eye-movements, for a period of approximately twenty (20) minutes. The conversation was then manually transcribed and the position in time of each word uttered by both the subject and experimenter was also manually recorded. This process resulted in a rich data set that allowed the precise overlay of eye-movements, in time with speech.

Fixations from the data sequence are selected as follows:

6) The current pupil position is detected and the corresponding 2D gaze angle is calculated.

7) If the subject is looking straight ahead, the frame is skipped. Otherwise, it is classified into one of the eight categories listed below, the class is recorded and a count is maintained.

8) On the next frame, if the classified position is the same as the one recorded, the count is updated. If the subject is now looking straight ahead, the count is reset.

9) If the count reaches a threshold, the frame is selected. This allows fixations to be selected.

10) When a frame is selected, is set to a lower threshold and the algorithm continues onto the next frame. This allows for search paths to be detected and its component frames selected.

Eye-movements are classified into eight (8) classes as per the classes introduced in NLP [2];

specifically up, up and to the left, up and to the right, left, down, down and to the left, down and

to the right, right. For a 2D gaze angle and, a class is assigned as follows:

–  –  –



Pages:     | 1 |   ...   | 17 | 18 || 20 | 21 |


Similar works:

«Political Psychology, Vol. 31, No. 4, 2010 doi: 10.1111/j.1467-9221.2010.00768.x The Behavioral Logic of Collective Action: Partisans Cooperate and Punish More Than Nonpartisansp ops_768 595.616 Oleg Smirnov Stony Brook University Christopher T. Dawes University of California, San Diego James H. Fowler University of California, San Diego Tim Johnson Stanford University Richard McElreath University of California, Davis Laboratory experiments indicate that many people willingly contribute to...»

«3rd Edition, 2009 Introduction to the 3rd Edition nd It has been just over a decade since the 2 Edition of the Mush with P.R.I.D.E. Sled Dog Care Guidelines were published. During that time, scientists have made great strides in their understanding of dog physiology, psychology and behavior. Researchers have studied working sled dogs, with the support of their mushers, some of whom are also Mush with P.R.I.D.E. members. Many of these research projects have validated sled dog care methods that...»

«Inga Jasinskaja-Lahti PSYCHOLOGICAL ACCULTURATION AND ADAPTATION AMONG RUSSIAN-SPEAKING IMMIGRANT ADOLESCENTS IN FINLAND Helsinki 2000 ii Sosiaalipsykologisia tutkimuksia Socialpsykologiska studier Social psychological studies Kustantaja / Publisher: Helsingin yliopiston sosiaalipsykologian laitos / Department of Social Psychology, University of Helsinki Toimituskunta / Editorial Board: Klaus Helkama, puheenjohtaja / chair person Karmela Liebkind Rauni Myllyniemi Anna-Maija Pirttilä-Backman...»

«ODYSSEY OFTHE MIND SPONTANEOUS PROBLEM SOLVING SPONTANEOUS PROBLEMS Verbal Hands -On Hands -On Verbal VERBAL In a Verbal spontaneous problem, the team is given a brainstorming-type problem to solve in a specific amount of time and scored according to he number and creativity of responses generated. The order in which members respond is usually random, and a higher point value is awarded for a creative answer than a common one. Examples: Name uses for a jack-o-lantern after Halloween. Name...»

«A WARHAMMER 40,000 NOVEL BLOOD PACT Gaunt’s Ghosts 12 (The Lost 05) Dan Abnett (An Undead Scan v1.1) For Dave Taylor It is the 41st millennium. For more than a hundred centuries the Emperor has sat immobile on the Golden Throne of Earth. He is the master of mankind by the will of the gods, and master of a million worlds by the might of his inexhaustible armies. He is a rotting carcass writhing invisibly with power from the Dark Age of Technology. He is the Carrion Lord of the Imperium for...»

«The International Journal of Indian Psychology ISSN 2348-5396 (e) | ISSN: 2349-3429 (p) Volume 3, Issue 2, No.9, DIP: 18.01.164/20160302 ISBN: 978-1-329-97719-8 http://www.ijip.in | January March, 2016 Impact of Yogic Practises on Risk Taking Behavior of Attension Deficit Hyperactivity Disorder Children Piyush Dubey1*, Garima Singh Kathait1, Dr. Anita Puri Singh 2 ABSTRACT Attention deficit hyperactivity disorder is one of the most common childhood disorders and can continue through adolescence...»

«Some Ways that Maps and Diagrams Communicate Barbara Tversky Department of Psychology, Stanford University Stanford, CA 94305-2130 bt@psych.stanford.edu Abstract. Since ancient times, people have devised cognitive artifacts to extend memory and ease information processing. Among them are graphics, which use elements and the spatial relations among them to represent worlds that are actually or metaphorically spatial. Maps schematize the real world in that they are two-dimensional, they omit...»

«Incidents in the Life of Madame Blavatsky by A.P. Sinnett Incidents in the Life of Madame Blavatsky by A.P. Sinnett Compiled from information supplied by her relatives and friends The Theosophical Publishing House, London 1913 AUTHOR'S PREFACE [Page 5] THE first edition of this book, published in 1886, was issued during Madame Blavatsky's lifetime as an indirect protest against the cruel and slanderous attack on her embodied in the Report to the Committee of the Psychical Research Society...»

«Le Mans (not just) for Dummies The Club Arnage Guide to the 24 hours of Le Mans 2015 Every input was pure reflex things were coming at me everywhere I looked. For about 50 percent of the lap I felt like I was on the verge of a massive accident. Mark Blundell commenting his pole position lap in a 1.100 hp Nissan Group C at Le Mans 1990 Copyright The entire contents of this publication and, in particular of all photographs, maps and articles contained therein are protected by the laws in force...»

«355 366. YTTERVALLA, FRUSTUNA SN, DAGA HD.366. Yttervalla, Frustuna sn, Daga hd. I en källare S om boningshuset i den östra gården i Yttervalla by finnas två mindre stycken av en runsten inmurade i källartaket. Källaren är enligt uppgift av gårdens ägare (herr Alb. Andersson) byggd år 1877. Stenmaterialet var delvis fört från en gammal grund i Långlid, delvis hopsamlat i backarna N om byn. Möjligen finnas i v'ággoch takmurarna flera rester av den sönderslagna runstenen, ehuru nu...»

«Chapter 11 COMMUNICABLE DISEASES ♦ Communicable Disease ♦ Reportable Diseases List ♦ Chickenpox ♦ Fifth Disease (Erythema Infectiosum) ♦ Infectious Hepatitis (Hepatitis A) ♦ Hepatitis B ♦ Impetigo ♦ Measles (Rubella, Red Measles, 10 day Measles, Hard Measles) ♦ Mononucleosis ♦ Mumps ♦ Pink Eye (Conjunctivitis) ♦ Pinworms ♦ Ringworm ♦ Rubella (German Measles, 3 day Measles) ♦ Scabies ♦ Scarlet Fever COMMUNICABLE DISEASE Communicable diseases are those diseases...»

«CHAPTER 1 Qualitative Research and Habits of Mind Story is far older than the art of science and psychology and will always be the elder in the equation no matter how much time passes. —Clarissa Pinkola Estes Women Who Run With the Wolves (1996) B ecause the researcher is the research instrument in qualitative research projects, it is important for the researcher to practice and refine techniques and habits of mind for qualitative research. Habits of mind in this text will include observation...»





 
<<  HOME   |    CONTACTS
2016 www.dissertation.xlibx.info - Dissertations, online materials

Materials of this site are available for review, all rights belong to their respective owners.
If you do not agree with the fact that your material is placed on this site, please, email us, we will within 1-2 business days delete him.