3D Partial scans models for people recognition with a RGB-D sensor
Research developed in this thesis is motivated by a vision of a future where social activities and personalized services are delivered in intelligent environments, entering a new era of more natural and wholesome human-machine interactions. While people are moving freely within these environments, i...
| Autor: | |
|---|---|
| Tipo de recurso: | tesis doctoral |
| Estado: | Versión publicada |
| Fecha de publicación: | 2020 |
| País: | España |
| Institución: | CBUC, CESCA |
| Repositorio: | TDR. Tesis Doctorales en Red |
| OAI Identifier: | oai:www.tdx.cat:10803/671232 |
| Acceso en línea: | http://hdl.handle.net/10803/671232 https://dx.doi.org/10.5821/dissertation-2117-342625 |
| Access Level: | acceso abierto |
| Palabra clave: | Àrees temàtiques de la UPC::Informàtica 004 68 |
| Sumario: | Research developed in this thesis is motivated by a vision of a future where social activities and personalized services are delivered in intelligent environments, entering a new era of more natural and wholesome human-machine interactions. While people are moving freely within these environments, individuals should be recognized by an artificial agent through a wide range of diverse tasks: detection, identification, re-identification and tracking. In this context, facial features are not enough and thus a need appears about finding new sources of descriptive human information to support those actions.Novel approaches in people recognition using human bodies as the main target are introduced throughout this dissertation. Our proposals intend to take full advantage of the powerful capabilities of one device only, regardless of whether the technology exploited is 2D (simple camera) or 3D (RGB-D sensor), favoring the effective use of minimum resources and combining less intensive measures. Applications start with the automatic landmarking of human shapes by learning a small dataset, upgraded by on-line 3D body shape contour tracking with an RGB-D sensor; next, groups of people are categorized and tracked in a public space and studied as a cognitive emulation of human behavior towards relationships generated through spatial and motion interactions; finally, a natural user interface is created with a RGB-D sensor for the identification and re-identification of individuals on the scene in real-time.By employing practical computer vision and machine learning techniques, experimental evaluation will show fair and adequate performance of the introduced systems, comparable to other more complex ones. All this implementation is carried out with due respect for human perception and the quality of their interaction with the intelligent agents. Always bearing in mind that their application goes hand in hand with the idea of eventually incorporating them in smart environments and robotic platforms. |
|---|