Tools
Tools
Pokhriyal, Mayank (2024) A Versatile Emotion Detection System: Independent Models for Image, Video, and Audio Analysis. Masters thesis, Dublin, National College of Ireland.
Preview |
PDF (Master of Science)
Download (1MB) | Preview |
Preview |
PDF (Configuration Manual)
Download (1MB) | Preview |
Abstract
Facial expressions are crucial aspects of human interpersonal communication and business that shape the sphere of healthcare and other professions, as well as human machines interactions. In this thesis, several methods are introduced based on deep learning for emotion recognition. The reported idea is to offer emotion detection services adaptable to the needs of the customer and the type of signal, whether image, live camera feed, video or audio. For image emotion recognition, several CNN architectures, including the VGGNet and ResNet50, were used and then were trained with transfer learning on large facial emotion databases. Techniques for augmenting the data set, tuning of its hyperparameters and class weight balancing were also employed to enhance the effectiveness of the above models.
In the context of detecting emotions from audio, Recurrent Neural Networks (RNNs), especially Long ShortTerm Memory (LSTM) networks were used for processing speech signals and for predicting the emotional states based on extracted acoustic parameters.
The performance of each model was separately assessed based on benchmark data set, however, to measure the functionality of the complete system, it was examined on live video stream and recorded media files. Thus, experimental evaluation reveals that the models produce satisfactory accuracy, precision, recall, and F1-score, which show good performance for various input modalities. This makes the proposed solutions flexible enough to enable users to select the most appropriate emotion detection approach for their given application environment, regardless of the context of use or the type of data that is being used. This work discusses the prospects and limitations of emotion recognition and offers a systematic approach to construct accurate, individual-focused emotion identification systems to be implemented in practical applications.
| Item Type: | Thesis (Masters) |
|---|---|
| Supervisors: | Name Email Simiscuka, Anderson UNSPECIFIED |
| Subjects: | Q Science > QA Mathematics > Electronic computers. Computer science T Technology > T Technology (General) > Information Technology > Electronic computers. Computer science H Social Sciences > HM Sociology > Information Science > Communication B Philosophy. Psychology. Religion > Psychology > Emotions Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Machine learning |
| Divisions: | School of Computing > Master of Science in Artificial Intelligence |
| Depositing User: | Ciara O'Brien |
| Date Deposited: | 20 Jun 2025 09:22 |
| Last Modified: | 20 Jun 2025 09:22 |
| URI: | https://norma.ncirl.ie/id/eprint/7959 |
Actions (login required)
 |
View Item |