ALS is a neurodegenerative illness that cause patients to lose
motor control resulting in a locked in syndrome in which patients retain
cognitive function but no longer have the ability to speak or write.
Because eye movement is typically retained into late stages of ALS,
eye gaze is the primary means of communication for this population.
High tech solutions using eye tracking technology
offer a communication solution but cost thousands of dollars.
Additionally, commercial eye trackers rely on infrared cameras
that do not perform well in some situations such as bright sunlight.
E-tran boards are a low cost, low tech solution
used by people who cannot afford expensive eye tracking systems or
used as a backup in settings where eye tracking technology fails.
A caregiver holds up a transparent board that has groupings of letters
and the person with ALS does two eye gestures to select a letter.
The first gaze gesture indicates a grouping and
the second indicates the position of a letter within the grouping.
Using E tran boards is slow and requires practice and skill for
interpreting gaze patterns.
We present a mobile phone application that automates
the E-tran board experience.
Providing a low cost alternative or
supplement to commercial eye tracking systems that is faster and
easier to use than low tech E-tran boards.
To use this system, the interpreter holds the phone and
points the back camera to the speaker with ALS.
A printed key taped to the phone's case
provides a visual indication of letter groupings for the speaker.
Upon first used their is a simple calibration procedure
requiring only a few seconds.
>> Calibration start please look up,
left, right, down, close,
open, calibration complete.
>> Our app, consists of three major components.
One, eye gaze recognition.
Two, a word prediction engine.
And three, a text entry interface.
For each camera frame our system first detects the presence of a face
and aligns line marks, extracts an image of each eye and
normalizes each eye image.
Then our system classifies eye gaze gestures
by matching the normalized eye image with our calibration gaze templates.
To simplify gesture recognition,
there are only four groupings of letters.
So each up, down, left or
right eye gaze is ambiguous in terms of which letter it may represent.
Similar to other ambiguous text entry systems like T9
our system interprets a series of consecutive gestures
to find all possible words that the sequence may represent.
And displays to the interpreter a rank in vestless
of likely words based on a language model.
In addition to moving the eyes up, down, left,
or right to enter a character, the speaker can
optionally wink their left eye to remove the last character or
wink their right eye to confirm the end of a word.
The interpreter sees all word predictions for
the current word as well as the already confirmed words
from this communication exchange on the phone screen.
The interpreter can speed up typing by selecting a word prediction based
on the context and can use agreed upon eye gestures to confirm this
prediction with speaker then add the work to the growing sentence.
Our app also offers a front facing camera mode which can be used when
the stand or
mount is available to attach the device to a user's wheelchair.
>> Next word, up, right,
up, and, how and.
Next word, down, left, down.
>> We also tested our app with ALS and
their communication partners to gain insights into its learnability and
usability by our target audience.
Study results, found that our apps significantly improve communication
time over standard E-tran boards contributing a new kind of portable,
low cost, usable communication solution for
people with sever motor disabilities.
Không có nhận xét nào:
Đăng nhận xét