Convolutional Neural Network-Based Gaze Estimation Using Inside-Out Camera

Chinsatit,  Warapon

WEKO3

lat lon distance

[[sub_check.contents]]

[[sub_radio.contents]]

Field does not validate

[[sub_attr.contents]]　

インデックスツリー

アイテム

Inside-Outカメラを用いた畳み込みニューラネットワークに基づく注視点推定

https://doi.org/10.18997/00007205

名前 / ファイル	ライセンス	アクション
jou_k_338.pdf (27.5 MB)

アイテムタイプ

学位論文 = Thesis or Dissertation(1)

公開日

2019-06-13

資源タイプ

資源タイプ識別子

http://purl.org/coar/resource_type/c_db06

資源タイプ

doctoral thesis

タイトル

Convolutional Neural Network-Based Gaze Estimation Using Inside-Out Camera

言語

タイトル

Inside-Outカメラを用いた畳み込みニューラネットワークに基づく注視点推定

言語

eng

著者

Chinsatit, Warapon

抄録

内容記述タイプ

Abstract

内容記述

The vision-based gaze estimation system （GES） involves multiple cameras, and such system can estimate gaze direction and what a user is looking at. The inside-out camera is the device to capture user eye and user vision. This system is widely used in many applications because the eye images with the pupil or cornea have much information. These applications have the capability to improve the quality of life of everyone especially a person with a disability. However, an end-user is difficult to access the ability of commercial GES device because of the high price and difficult to use. The budget GES device can be created with a general camera. The common method to estimate the gaze point from the vision-based GES is detected the pupil center position. However, the human eye has variable characteristics and the blinking makes reliable pupil detection is a challenging problem. The state-of-the-art method for the pupil detection is not designed for the wearable camera, the designed for the desktop/TV panels. A small error from the pupil detection can make a large error on gaze point estimation. This thesis presents the novel robust and accurate GES framework by using the learning-based method. The main contributions of this thesis can be divided into two main groups. The first main contribution is to enhance the pupil center detection by creating an effective pupil center detection framework. The second contribution of this thesis is to create the calibration-free GES. The first contribution is to enhance the accuracy of the pupil detection process. Handcraft and learning-based method are used to estimate the pupil center position. We design the handcraft method that using the gradient value and RANSAC ellipse fitting. The pupil center position was estimated by the proposed method and com-pared with the separability filter. The result shows the proposed method has a good performance in term of accuracy and computation time. However, when the user closes the eye, no eye present in the image, or a large unexpected object in the image, the accuracy will be decreased significantly. It is difficult for handcraft method to achieve good accuracy. The learning-based method has the potential to solve the general problem that becomes the focus of this thesis. This thesis presents the convolutional neural network （CNN） model to estimate the pupil position in the various situations. Moreover, this model can recognize the eye states such as open, middle, or closed eyes. The second contribution is to create the calibration-free GES. The calibration process is the process to create the coordinate transfer （CT） function. The CT function uses for transfer the pupil position to the gaze point on-scene image. When the wearable camera moves during the use case, the static CT function cannot estimate the gaze point accurately. The learning-based method has a potential to create a robust and adaptive CT function. The accurate calibration-free system can raise the accuracy of the GES. Furthermore, it makes the GES easy easier to use. We designed the CNN framework that has the ability to estimate the gaze position in the various situations. This thesis also presents the process to create the reliable dataset for GES. The result shows that proposed calibration-free GES can estimation the gaze point when glasses are moved.

言語

内容記述タイプ

TableOfContents

内容記述

1 Introduction||2 Pupil Detection using handcraft method||3 Convolutional neural network|| 4 Pupil detection using CNN method||5 Calibration free approach for GES||6 Character input system||7 Conclusion

備考

内容記述タイプ

Other

内容記述

九州工業大学博士学位論文学位記番号：情工博甲第338号学位授与年月日：平成31年3月25日

キーワード

主題Scheme

Other

主題

Gaze estimation system

キーワード

主題Scheme

Other

主題

Inside-out camera

キーワード

主題Scheme

Other

主題

Eye state classification

キーワード

主題Scheme

Other

主題

Convolutional neural network

キーワード

主題Scheme

Other

主題

Calibration-free gaze estimation system

アドバイザー

齊藤, 剛史

学位授与番号

甲第338号

学位名

博士（情報工学）

学位授与年月日

2019-03-25

学位授与機関

学位授与機関識別子Scheme

kakenhi

学位授与機関識別子

17104

学位授与機関名

九州工業大学

学位授与年度

内容記述タイプ

Other

内容記述

平成30年度

出版タイプ

VoR

出版タイプResource

http://purl.org/coar/version/c_970fb48d4fbd8a85

アクセス権

open access

アクセス権URI

http://purl.org/coar/access_right/c_abf2

ID登録

10.18997/00007205

ID登録タイプ

JaLC

戻る

views

See details

	Views

Versions

Ver.1

2023-05-15 12:53:43.328303

Show All versions

Cite as

Other

エクスポート

OAI-PMH

JPCOAR 2.0
JPCOAR 1.0
DublinCore
DDI

Other Formats

インデックスリンク

インデックスツリー

アイテム

Inside-Outカメラを用いた畳み込みニューラネットワークに基づく注視点推定

× Chinsatit, Warapon

Versions

Share

Cite as

Other

エクスポート

コミュニティ

メニューを最小化

インデックスリンク

インデックスツリー

アイテム

Inside-Outカメラを用いた畳み込みニューラネットワークに基づく注視点推定

× Chinsatit, Warapon

Versions

Share

Cite as

Other

エクスポート

コミュニティ