Affective Human Machine Interaction
| Project/Area Number |
15500141
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (C)
|
| Allocation Type | Single-year Grants |
|---|
| Section | 一般 |
|---|
| Research Field |
Sensitivity informatics/Soft computing
|
|---|
| Research Institution | University of Aizu |
|---|
Principal Investigator |
BERTHOUZE Nadia University of Aizu, Computer Software Department, Assistant Professor, コンピュータ理工学部, 講師 (20325971)
|
|---|
| Project Period (FY) |
2003 – 2004
|
| Project Status |
Completed (Fiscal Year 2004)
|
| Budget Amount *help |
¥2,900,000 (Direct Cost: ¥2,900,000)
Fiscal Year 2004: ¥800,000 (Direct Cost: ¥800,000)
Fiscal Year 2003: ¥2,100,000 (Direct Cost: ¥2,100,000)
|
|---|
| Keywords | Affective computing / Affective posture recognition / User modeling / Neural Networks / Cross-cultural studies / Non verbal communication / Categorization process / 感情を表すジェスチャの分類 / 追加学習 / 社会性ロボット / 感情の身体表現 / 感情を表す姿勢の記述 / CALMネットワーク / ヒューマンロボットコミュニケーション |
|---|
| Research Abstract |
The goal of the project was threefold
Gesture description: To support an automatic recognition process, we proposed 24 static posture descriptors characterizing postures in terms of the space occupied by the body, and the tendency of the gesture and 9 dynamic features characterizing the direction of the motion. The saliency of these posture features in discriminating between affective states was confirmed using various statistical methods, and in particular, discriminant analysis. Given the dearth of studies on affective gestures/postures in the fields of affective computing and Kansei engineering, this formal result was a very significant finding.
Gesture recognition system: We implemented a categorization neural network, based on Categorizing and Learning Modules (CALM), that incrementally learns to discriminate between affective gestures by integrating both supervised and unsupervised learning mechanisms. Following studies in neuroscience, we modified the topology of the neural networ
… More
k to handle horizontal interaction between the postures' dynamic and the static descriptions at a layer higher layer than the input one. Extensive studies have been carried out to tune and test the robustness of the system. The network was trained on 183 affective gestures displaying 4 different emotions (angry, fear, happy and sad). It was tested on 30 sets of 183 gestures obtained by varying the amount of noise added to the training set. The average classification performance was 80%, much higher than the other systems in the literature (Coulson, 2004), (Picard, 2004).
Cultural differences in affective posture recognition: Experiments involved 60 subjects from different cultural background: Japanese, Sri Lankan and Caucasian American.While there were no statistically meaningful differences between cultures in terms of how they perceived and rated fear-related postures, we could observe significant cultural differences in the intensity rating of angry, happy and sad.. With respect to our study on e-learning, this is a critical finding because it demonstrates that the affective state of a particular body posture should be differentially rated depending on the culture of the subject. Less
|
Report
(3 results)
Research Products
(24 results)
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
[Journal Article] Learning to Recognize Affective Body Posture2003
Author(s)
N.Bianchi-Berthouze, T.Fushimi, M.Hasegawa, A.Kleinsmith, H.Takenaka, L.Berthouze
-
Journal Title
IEEE International Symposium on Computational Intelligence for Measurement Systems and Applications(Lugano, Switzerland) July 29-31
Pages: 193-198
Description
「研究成果報告書概要(欧文)」より
Related Report
-
-
-
-
-
