Enhanced Visual Field Analyzer

Abstract
“Visual Field Analyzer” or VFA is a device used by ophthalmologist for mapping a patient’s field of vision. In a typical test the patient has to fix his gaze to a certain point in the middle of a test bowl (see picture). Then LEDs are onset for a short time at different locations on the test bowl. The patient must push a button each time he recognizes an onset light (LED). The software of the equipment automatically build a mapping of the areas which the patient could recognize.
In this project we propose a different method which enables to replace the push button by using reliable image processing software which tracks the patient’s eye movements. The patient is asked to look directly into the onset LED every time he recognizes an enlightened LED rather than starring to a fixed point in the middle of the test bowl. When the LED is offset he must return his gaze to the original central point.
The solution we propose is based on tracking the eye movement and estimating the correlation between the tracking information and the leds being switched on.
Image Processing
Finding the pupil center is a well known problem in the field of eye tracking, however in order to get good results from a wide variety of people and eye positions, we had to overcome many technical problems. The application that we developed uses a recursive algorithm based on thresholding. (For details see the full project report below).
The below images show some examples of our results. The red “”PLUS”” marks the central point of the pupil as found by our algorithm. Visually examined it is clear from the examples that the location of the pupils is recognized accurately for very different events, i.e. huge / small pupil open / closed eye / different types of eyes.

Data Processing
The stream of pupil center coordinates is carefully processed:

First, In order to solve the problem of different people having different geometric eye sizes, the coordinates are translated into a generic space, in which it is possible to compare all the subjects. In order to achieve that, a calibration factor is calculated for every new subject:

Each subject has certain ” t” coordinates (which are taken from the image space) and may differ due to different eye geometrics. Finding a factor which connects these coordinates to the actual physical coordinates on the VFA is achieved via calibration.
Second, The Normalized data is then filtered and we are left with only relevant data. (without Eye Blinks or Eye Movements)

Finally, we compare the filtered data and the area tested for, a measure is then produced for the correlation between the two:
A high correlation is translated as a positive ID ( A button pressed on).
A low correlation is translated as a negative ID ( A button not pressed on).

Results

In our tests 10 different LEDs were used at different angles and distance from the central point. 14 patients were tested. From a total of 140 tested LEDs 136 were recognized appropriate to the push button. This means a success rate of 97%. If we recheck each tested LED several times we might achieve almost 100% of success.

Tools
The Project was developed with MATLAB 6.5 and is compatible with an OPTOPOL PTS 900 Visual Field Analyzer.
The Package includes a reverse-engineered GUI that controls the PTS 900 manually.

Conclusions

1. The Test Procedure can now be activated without the button. The results we achieved are above 97% success.

2. A major advantage of the method developed is that instead of the two binary state that a button allows, here one can receive an additional indication regarding the duration and quality of a given ID. From our experience after a certain amount of time a person sees even his “blind spot”, in the old test he would have simply pressed the button. Now we have additional information that indicates a weak spot, even though it was seen.

3. It is possible that the fatigue problems that are encountered in the old test, where a person is asked to keep gazing forward even though a led has been switched on, reduce in the new test. We find it a bit more natural to gaze at the led as it is being recognized.