| 研究課題/領域番号 |
18K18168
|
|---|
| 研究機関 | 岡山大学 |
|---|
研究代表者 |
Yucel Zeynep 岡山大学, 自然科学研究科, 特任助教 (20586250)
|
|---|
| 研究期間 (年度) |
2018-04-01 – 2021-03-31
|
| キーワード | E-learning / Attention / Engagement / Behavior |
|---|
| 研究実績の概要 |
We performed a set of experiments to collect a video data set from several participants performing e-learning tasks with varying mental workload. On this footage, we detected eye blinks based on facial landmarks. We computed four features as frequency and duration of blinks, eye size and aspect ratio. We obtained promising results indicating that these features are in considerable correlation with apparent levels of engagement.
In addition, we made another set of experiments displaying emotionally stimulating visual content and collected electro-dermal activity data. We proved that humans' responsiveness decreases over time and can be estimated through an exponential attenuation model. Our model gave better results than deep neural networks when trained with the same amount of data.
|
| 現在までの達成度 (区分) |
現在までの達成度 (区分)
2: おおむね順調に進展している
理由
We performed two sets of experiments at the scheduled time. We also examined the data set on time and achieved some promising results, which help us plan the details of the next stages of the project.
|
| 今後の研究の推進方策 |
The outcomes of our study so far indicate that several blink and non-blink features have considerable correlation with level of engagement. This suggests that they can be used to estimate in a probabilistic manner the level of engagement. Thus we are considering to (i) derive the density distribution of these features for various levels of engagement for computing the likelihood of being engaged, (ii) blend together the likelihood values to integrate all available information, and (iii) evaluate the final estimation accuracy. Next year, we will pursue working towards these goals.
|
| 備考 |
研究代表者ホームページ
https://yucelzeynep.github.io/
|
