FAIRSpectra - Enabling the FAIRification of Analytical Science
Presentation yamin
1. iShadow: Design of a
wearable, real-time mobile
gaze tracker
Presenter: Yamin Tun
Addison Mayberry, Pan Hu, Benjamin
Marlin, Christopher Salthouse, Deepak
Ganesan (UMass)
12. Limitations/Critique
Varying performance for different users
Different depth of view for different faces
Visibility of the entire eye
Glasses placement
Obstructing user’s view
Imager + motion sensor
Gaze on same object with head movement
Editor's Notes
What is Gaze Tracking?
Gaze Tracking is measuring gaze points- estimating where a person is looking at in the physical world using eye movement information
Applications- Continuous real-time gaze tracking can be valuable/useful in a lot of applications such as
detection of unsafe behavior- such as tracking driver’s gaze and detect whether driver has attention on the road- prevent accidents- pedestrian and cars
Detecting psychological behaviors-
Pupil dilation- excitement
Pupil micromovements-detect fatigue
Power hungry, computation intensive device -> bulky
wearable devices
Easy Comp Vision problem- detect circle or the darkest patches- probably using some powerful MCU
-real-time
-power
- Google glasses- power consuming, as a wearable device- not realistic to run out of battery all the time
- but to do it in real-time with limited computation power- challenge
-right now, glasses use eye-facing camera at the back
STM32 Arm cortex- MCU
Stonyman- grayscale imager
Storage
How the eyeglasses work?
-feed in Eye images from back camera-> Output=gaze estimate
-Neural network to estimate gaze
->to build it, offline-training
~3 degree of error on average= gaze estimate error