Emotion and Movement Analysis Study from Asian and European Facial Expressions
- Resource Type
- Conference
- Authors
- Kayi, Berk; Erbasi, Zeynep; Ozmen, Samet; Kulaglic, Ajla
- Source
- 2023 3rd International Conference on Innovative Research in Applied Science, Engineering and Technology (IRASET) Innovative Research in Applied Science, Engineering and Technology (IRASET), 2023 3rd International Conference on. :1-5 May, 2023
- Subject
- Communication, Networking and Broadcast Technologies
Computing and Processing
General Topics for Engineers
Signal Processing and Analysis
Training
Emotion recognition
Computational modeling
Face recognition
Data models
Internet
Task analysis
movement detection
Asian Facial Expression
European Facial Expression
- Language
The proposed work is done incollaboration with the different partners under the EUREKA project in order todevelop a 360-degree behavioral assessment product that enables theidentification, extraction and evaluation of facial expression and gesturalbehaviors through the recognition of human emotions via video recordings. Thecollected videos are preprocessed and transformed into frames from whichethnicity of person is determined. Based on the ethnicity two different machinelearning algorithms are applied for detecting the dominant emotion in video.The models use pre-trained model trained using benchmark data sets, FER-2013and CK+; and images collected by Google search and videos collected during themeetings and presentations are used to improve the accuracy of pre-trainedmodels. More than 292680 images and 60 videos are used for training and testingperformances. The Residual Masking Network and CNN architectures are used aspre-trained models. The accuracy of proposed solutions is 75.2% for Asianpersons and 86.6% for Caucasian persons. Implementing the proposed solutioninto real sector and collecting the more data the accuracy will be improved.