| 研究実績の概要 |
We decided to build empirical models for the four behavioral variables derived from facial landmarks. To that end, we used kernel density estimation. In addition, we considered the extremities of engagement level (fully engaged and completely disengaged) as benchmark states. In that respect, we obtained 8 models corresponding to the pair of each of the four behavioral variables and two benchmark states.
We then proposed a probabilistic method for estimating level of engagement. In that respect, each feature value is evaluated in its two corresponding models representing the benchmark states. Making use of the fact that these are complementary, we computed the probability of engagement.
Our results indicate that, the features provide considerable accuracy for distinguishing the two (benchmark) states.
|
| 今後の研究の推進方策 |
At the current stage, we handle each behavioral feature in an isolated manner. From now we will search for possibilities to incorporate the evidence obtained from each single feature into a single outcome. To that end, we will investigate the independence of features and build joint distributions, if necessary.
|
