3D Human pose estimation from egocentric inputs
Egocentric pose estimation is essential for developing embodied AI systems capable of interacting naturally with humans and their environments. This thesis addresses the challenges of first-person pose estimation through a series of interconnected studies. The first study, BoDiffusion, presents a ge...
| Autor: | |
|---|---|
| Tipo de recurso: | tesis doctoral |
| Estado: | Versión aceptada para publicación |
| Fecha de publicación: | 2024 |
| País: | Colombia |
| Institución: | Universidad de los Andes |
| Repositorio: | Séneca: repositorio Uniandes |
| Idioma: | inglés |
| OAI Identifier: | oai:repositorio.uniandes.edu.co:1992/75400 |
| Acceso en línea: | https://hdl.handle.net/1992/75400 |
| Access Level: | acceso abierto |
| Palabra clave: | Egocentric vision Pose estimation Pose forecasting Ingeniería |
| Sumario: | Egocentric pose estimation is essential for developing embodied AI systems capable of interacting naturally with humans and their environments. This thesis addresses the challenges of first-person pose estimation through a series of interconnected studies. The first study, BoDiffusion, presents a generative model that synthesizes full-body motion from sparse inputs. The second study, Ego-Exo4D, establishes a benchmark for pose estimation in real-life settings with diverse activities. The final study, EgoCast, focuses on current pose estimation and forecasting in the wild, integrating visual and proprioceptive inputs to handle dynamic and unscripted environments. Together, these contributions provide robust, temporally consistent methods for real-world 3D pose estimation. |
|---|