2014 Fiscal Year Final Research Report
Gaze estimation by image matching with eye-ball model CG
| Project/Area Number |
25560105
|
|---|
| Research Category |
Grant-in-Aid for Challenging Exploratory Research
|
| Allocation Type | Multi-year Fund |
|---|
| Research Field |
Educational technology
|
|---|
| Research Institution | Tokyo Institute of Technology |
|---|
Principal Investigator |
|
|---|
| Research Collaborator |
TAKAHASHI Shota
HASUMI Kozue
|
|---|
| Project Period (FY) |
2013-04-01 – 2015-03-31
|
| Keywords | 視線推定 / 画像処理 / 眼球モデル |
|---|
| Outline of Final Research Achievements |
We proposed and developed an eye-gaze estimation system from a human face image shot by a normal camera.Gaze direction is estimated by adopting parameters that minimize error between an eye image and a CG image. The eye region is trimmed by using human face feature points. The CG image has angles of eye-ball rotation and openness of eyelids as the parameters. The best parameters are searched so that the zero-mean normalized cross-correlation between the shot image and the CG is minimized.
In the experiment, subject's head were fixed first. Average estimation errors are not very different when subjects gaze upward, but when gazing horizontal direction and downward, the errors are made small by using the eyelid model. Next, we conducted experiment that allow subjects to freely move their heads. Gaze direction must be calculated using all the parts of the proposed system. In this case average estimation error became large. Individual differences seem to reduce the accuracy.
|
| Free Research Field |
信号処理、教育工学
|
