AI-based construction of 3D human face representations from 2D images for emotion recognition
The commercialization of antennas that would operate in an extreme-frequency band is expected to be available in the following decade, which will provide a high- transmission rate integrated into small compact devices. The current restrictions the EU is posing on the machine learning techniques for...
| Autor: | |
|---|---|
| Tipo de recurso: | tesis de maestría |
| Fecha de publicación: | 2024 |
| País: | España |
| Institución: | Universitat Politècnica de Catalunya (UPC) |
| Repositorio: | UPCommons. Portal del coneixement obert de la UPC |
| Idioma: | inglés |
| OAI Identifier: | oai:upcommons.upc.edu:2117/406658 |
| Acceso en línea: | https://hdl.handle.net/2117/406658 |
| Access Level: | acceso abierto |
| Palabra clave: | Human face recognition (Computer science) Artificial intelligence extremely high frequency facial emotion recognition monocular depth estimation facial reconstruction freqüència extremadament alta reconeixement d'emocions facials estimació de la profunditat monocular reconstrucció facial Reconeixement facial (Informàtica) Intel·ligència artificial Àrees temàtiques de la UPC::Informàtica::Intel·ligència artificial |
| Sumario: | The commercialization of antennas that would operate in an extreme-frequency band is expected to be available in the following decade, which will provide a high- transmission rate integrated into small compact devices. The current restrictions the EU is posing on the machine learning techniques for facial emotion recognition force us to find ways to isolate the identity of a subject from the actual expression, making the approach of capturing through wireless signals suitable for that. The FER applications are well exploited in a wide range of areas, from the creation of avatars in computer graphics, to physiological analysis for diagnostic purposes in precision medicine. At the best of our knowledge, this work is focused on the analysis on the 2D RGB images (which despite the good results, there are still challenges to solve for wild type of data) leaving the geometry processing on point cloud/depth images scope still limited in terms of facial gesture detection. The work presented on this thesis aims to set the basis for the validation of the FER applications that utilize a point cloud or depth information as an input, by an- alyzing the current state-of-the-art from among different areas like depth estimators, face reconstruction and 2D RGB image facial emotion recognition, giving a common knowledge and presenting results across these to tackle this goal. |
|---|