Research study on eye gaze estimation using neural network
Since the 1800s, studies of eye movement and eye tracking has long attracted interest from various researchers and scientists. However, eye tracking technology has slowly moved towards eye gaze estimation technology instead as it has been proven to be more useful in real-world applications. Th...
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Final Year Project |
| Language: | English |
| Published: |
2019
|
| Subjects: | |
| Online Access: | http://hdl.handle.net/10356/77032 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | Nanyang Technological University |
| Language: | English |
| Summary: | Since the 1800s, studies of eye movement and eye tracking has long attracted interest from
various researchers and scientists. However, eye tracking technology has slowly moved
towards eye gaze estimation technology instead as it has been proven to be more useful in real-world applications.
This project aims to collect eye gaze data, explore methods for eye gaze estimation using
convolutional neural network and implement the methods on mobile devices to explore the
application of the eye gaze estimation technology. Neural networks are a part of deep learning
technology which has been adopted by various major IT companies around the globe. It
utilizes artificial neurons to learn the patterns that exist in complex data such as image and
audio to predict certain results.
To collect eye gaze data, two Android applications, GazeCollect and Gazestimate was
developed to collect 2 different types of eye gaze data. GazeCollect gathers data of participants
looking at random red dots displayed on the mobile device. On the other hand, Gazestimate
gathers data through a simple card matching game which participants are required to complete.
For this project, classification method was explored in this project to predict eye gaze location
on mobile devices by segmenting the mobile device’s screen into 32 areas. A combination of
image and scalar inputs are used as input for the convolutional neural network model to
compensate for the difference in device orientation and size. Resnet is used in the
convolutional network model architecture as Resnet has been proven to perform accurately
in image recognition task. The classification method explored yielded a best accuracy of 47.76
to 49.37% accuracy on predicting eye gaze location.
Finally, Gazestimate simple card matching game was modified to be interactive using eye gaze
estimation. Additional functionality was added to Gazestimate to display the exact eye gaze
prediction on the device’s screen by displaying red dots too. These functionalities were used
to showcase eye gaze estimation techniques during Nanyang Technological University’s Open
House 2019 |
|---|