Active Reconstruction of Dynamic Scenes Using Drones
Event details
| Date | 11.06.2018 |
| Hour | 11:00 › 13:00 |
| Speaker | Sena Kiciroglu |
| Location | |
| Category | Conferences - Seminars |
EDIC candidacy exam
Exam president: Prof. Bertrand Merminod
Thesis advisor: Prof. Pacal Fua
Co-examiner: Dr. Ronan Boulic
Abstract
Active reconstruction of dynamic scenes has become an interesting direction of research with the rising popularity of commercially available drones. The key idea is to actively reposition a drone during recording so that the performance of the task being performed is improved. We focus in particular on the task of 3D human pose estimation. A possible example application could be a personal ski trainer, where the drone follows a skier and uses its camera images to estimate their pose. According to the pose estimation results, the drone will reposition itself, performing active reconstruction. In this talk, three papers will be discussed. The first paper focuses on 3D human pose estimation. The second paper is about 3D human model reconstruction from video footage. The third paper is on active reconstruction of static scenes. We will see that using ideas from all three papers, we can form a novel and exciting direction for new research.
Background papers
Tracking People with Twists and Exponential Maps, by C. Bregler and J. Malik
Video Based Reconstruction of 3D People Models, by T. Alldieck, et al.
An Information Gain Formulation For Active Volumetric 3D Reconstruction, by S. Isler, et al.
Exam president: Prof. Bertrand Merminod
Thesis advisor: Prof. Pacal Fua
Co-examiner: Dr. Ronan Boulic
Abstract
Active reconstruction of dynamic scenes has become an interesting direction of research with the rising popularity of commercially available drones. The key idea is to actively reposition a drone during recording so that the performance of the task being performed is improved. We focus in particular on the task of 3D human pose estimation. A possible example application could be a personal ski trainer, where the drone follows a skier and uses its camera images to estimate their pose. According to the pose estimation results, the drone will reposition itself, performing active reconstruction. In this talk, three papers will be discussed. The first paper focuses on 3D human pose estimation. The second paper is about 3D human model reconstruction from video footage. The third paper is on active reconstruction of static scenes. We will see that using ideas from all three papers, we can form a novel and exciting direction for new research.
Background papers
Tracking People with Twists and Exponential Maps, by C. Bregler and J. Malik
Video Based Reconstruction of 3D People Models, by T. Alldieck, et al.
An Information Gain Formulation For Active Volumetric 3D Reconstruction, by S. Isler, et al.
Practical information
- General public
- Free
Contact
- EDIC - [email protected]