Body-centric understanding of 3D environments
Abstract/Contents
- Abstract
- Human environments are designed by people to facilitate human actions. Viewing environments as spaces for action motivates the use of an action representation to factor and explain the structure and function of environments. This dissertation proposes to encode human actions as coupled configurations of the body with its surrounding environment in a body-centric representation. By joining these two aspects of human interaction we can analyze 3D scenes through the lens of action, and jointly generate 3D scenes and human poses depicting actions. The body-centric view proposed by this dissertation treats body parts and parts of objects as primitives linked through physical contact, gravitational support, and visual attention in a graph characterizing a given interaction. These primitives and links are encoded in an interaction graph capturing features of the pose configuration and the geometry of the objects present during an action. By recording real-world observations of common actions and converting them into this representation we can obtain aggregate statistics of pose and object configurations under different actions which we summarize in a prototypical interaction graph (PiGraph). The PiGraphs model formalizes a body-centric view of interactions and can be learned directly from observational data captured by commodity RGB-D sensors. PiGraphs can be used to predict the likelihood of common actions occurring over the space of an input 3D environment. They also enable action-based 3D scene synthesis resulting in higher perceived plausibility and quality compared to prior approaches in scene synthesis. Finally, by connecting PiGraphs to natural language we can create an end-to-end text-to-interaction system that jointly generates a 3D scene and human pose depicting an interaction described in natural language.
Description
| Type of resource | text |
|---|---|
| Form | electronic; electronic resource; remote |
| Extent | 1 online resource. |
| Publication date | 2016 |
| Issuance | monographic |
| Language | English |
Creators/Contributors
| Associated with | Savva, Manolis | |
|---|---|---|
| Associated with | Stanford University, Department of Computer Science. | |
| Primary advisor | Hanrahan, P. M. (Patrick Matthew) | |
| Thesis advisor | Hanrahan, P. M. (Patrick Matthew) | |
| Thesis advisor | Agrawala, Maneesh | |
| Thesis advisor | Guibas, Leonidas J | |
| Advisor | Agrawala, Maneesh | |
| Advisor | Guibas, Leonidas J |
Subjects
| Genre | Theses |
|---|
Bibliographic information
| Statement of responsibility | Manolis Savva. |
|---|---|
| Note | Submitted to the Department of Computer Science. |
| Thesis | Thesis (Ph.D.)--Stanford University, 2016. |
| Location | electronic resource |
Access conditions
- Copyright
- © 2016 by Manolis Savva
- License
- This work is licensed under a Creative Commons Attribution Non Commercial 3.0 Unported license (CC BY-NC).
Also listed in
Loading usage metrics...