-
Material properties will require to combine bilineal and quadratic projection modes of lightning
- It should be done only if possible as a minor contribution
-
Real Time Camera Tracking will be approximated using motion estimation
-
The PhD is focused on monocular video but RGB-D cameras are not discarted yet
-
Long loop scenes
- Using tangent space from skew-symmetric matrices of Lie algebra
- Open loops can be closed by interpolating using a bilineal transformation
-
Mesh simplification during reconstruction
- Reduces memory footprint
- Boost future computations
- Determine prototypes and software requirements
- Design pipeline architecture
- Establish criteria for a quantitative evaluation of the results using percentage of error or improvement from state of the art
- Tracking and Mapping will be two coupled components with a closed feedback between them
- Noise from measures must be taken into account in every step: minimize the reprojection error
- Consider VTK or some visualization library for point clouds
- Energy functional and evaluation / validation procedures require more explanation
- Consider Wallenstein distance: it looks for similarities between
grouped data by mesuring orden
$k$ moments- It includes the Hausdorff measure (2010)
- Consider background segmentation to reduce the noise in the scene
- In the evaluation, it is recommended to parametrize the total error wrt camera velocity and background distance
- First implementation of a 3D reconstruction pipeline
- Using FAST features
- KLT tracking
- Triangulation
- Describe the smart meshing procedures taken into consideration