[Work Log]
June 17, 2013
Continuing work on foreground curve model. Editing correspondence/clean_correspondence2.m.
Logic overview
- smooth triangulated points & estimate their indices
- fill-in untriangulatable points/indices
- resolve many-to-one data-to-3d-curve correspondences by picking best match
- triangulate each data point against 3d curve to get mean and covariance for point
Logic detail
1. estimate indices by chord-length parameterization
2. repeat N times:
2.1 smooth triangulated points using current index estimate (posterior mean approx.)
2.2 re-estimate indices using chord-length parameterization
3. for each untriangulatable point
3.1 triangulate against cubic interpolation of neighboring points (Newton's method)
3.2 store resulting point and index with smoothed points from 2.1
4. for each 2D data curve
4.1 triangulate against corresponding 3d point in smoothed curve to get likelihood mean
4.2 compute curvature at this point to get likelihood precision
03:13:30 PM
- First run, addressing syntax errors.
- Plotting first test: triangulated curve w/ and wo/ gap-filling
- Issue: gap-fill at beginning maps exactly to first non-gap point
- Issue: interior gap point falls exactly on non-gap point
- need to merge points?
04:06:03 PM wrapping up...
TODO
- diagnose coincident-point issue -- bug or not?
- Finish testing - plot per-view likelihood points
- Add point-merging