RoboSWOP Meeting Nov 23
Location
David's Office
Agenda
- Review action items of last meeting #4
- Review the space carving concept that Emil will explain
- Discuss next steps
Participants
Event selection concept
- Start from an event E in a starting catalogue C_1 (with parameters {x_1i}_i), e.g., the CACTUS catalogue.
- For each of the other catalogues C_j (listed with parameters {x_ji}_i), calculate approximations {x_ji_approx}_i for (some of) the parameters {x_ji}_i, starting from the known parameters {x_1i}_i. Let there be N such catalogues.
- For each of these catalogues C_j, select all events {E_jk}_k which have parameters {x_ji}_i sufficiently close to {x_ji_approx}_i. The "sufficiently close" is defined by broad acceptance criteria such as limited time difference and limited difference in e.g., longitude and latitude. Let {E_jk}_k be the list of such events in catalogue C_j. The results of this is a set of events (E, E_1k, E_2k, …, E_Nk) all "sufficiently close" to the starting event E.
- (optional step): Consider all possible subsets of the master set of all accepted events (E, E_1k, E_2k, …, E_Nk), for each subset, define a "score" how coherent the subset is. For example E (say a CME) might be well connected to E_2k (say a flare) and to E_3K (say a dimming), but if E_2K and E_3k are not well connected mutually (eg location-wise) then the subset (E, E_2k, E_3k) would not be coherent and therefore get a low score. The following step considers either the full combination (E, E_1k, E_2k, …, E_Nk) or only the highly-scored subsets of that.
- Employ the space carving results to decide which combinations (E, E_1k, E_2k, …, E_Nk) possibly constitute a meta-event, and rank the combinations from highest to lowest probability if there is more than one such combination.
- The final characterization of the meta-event is given by the space carving results.
Voxel/Space/Silhouette Carving - Volume Intersection
A list of techniques @emilk needs to explore; See the c=2 example here: http://faculty.cs.tamu.edu/jchai/CSCE_CP/lecture10.ppt (reminder; we’re doing binary, not using color/luminance information, also no feature matching between viewpoints,…)
Basic algorithm.
- For each voxel (e.g., 1000 x 1000 x 1000, spanning 40Mkm^3 ), For each camera view, If voxel lies ‘inside’ a binary CME blob (extrapolated from the viewpoint), color it black.
- This means we create a 1000 x 1000 x 1000 - 3D space for each viewpoint, and then intersect them.
More stuff to study.
-
A Theory of Shape by Space Carving http://citeseerx.ist.psu.edu/viewdoc/download;jsessionid=7BDC373D0EB92A6E35EBA2F24EBBF469?doi=10.1.1.163.250&rep=rep1&type=pdf
-
Real-Time GPU-based Voxel Carving with Systematic Occlusion Handling https://pdfs.semanticscholar.org/8d50/0c61bf9fa6b16fce1f407bc2174679d3c120.pdf
-
Hardware Accelerated Voxel Carving http://gram.eng.uci.edu/comp.arch/new_pubs/c78.pdf
-
Python space carving https://github.com/LG95/Space-carving
-
MATLAB carving a dinosaur https://github.com/matheusportela/3D-reconstruction/blob/master/code/carving.m https://nl.mathworks.com/matlabcentral/fileexchange/26160-carving-a-dinosaur
-
Old school voxel carving https://nghiaho.com/?p=2124
-
C++ Voxel Carving example https://github.com/xocoatzin/Voxel-Carving
-
C++ Voxel Carving example; https://github.com/subokita/Sandbox/tree/master/VoxelCarving/VoxelCarving
-
More generally, intersection of shapes.. https://pcjericks.github.io/py-gdalogr-cookbook/geometry.html http://pro.arcgis.com/en/pro-app/tool-reference/3d-analyst/intersect-3d-3d-analyst-.htm https://pypi.org/project/frentos/
-
How to display the resulting collection of voxels. (JHelioviewer..?)
Searching for (python) libraries that can do this… OpenCV?
Emil will ask Marilena for a handful of good CME that she has studied before. Then collect Solar Demon dimming, flare and CACTus CME detections for this.