Download - CSE590 V : Multi-View Reconstruction
![Page 1: CSE590 V : Multi-View Reconstruction](https://reader035.vdocuments.site/reader035/viewer/2022062410/568163e4550346895dd542fb/html5/thumbnails/1.jpg)
CSE590 V : Multi-View Reconstruction
Avanish Kushal
![Page 2: CSE590 V : Multi-View Reconstruction](https://reader035.vdocuments.site/reader035/viewer/2022062410/568163e4550346895dd542fb/html5/thumbnails/2.jpg)
Multi-View Reconstruction Preserving Weakly-Supported Surfaces [CVPR 2011]
Michael JancosekTomas Padjla
![Page 3: CSE590 V : Multi-View Reconstruction](https://reader035.vdocuments.site/reader035/viewer/2022062410/568163e4550346895dd542fb/html5/thumbnails/3.jpg)
Motivation
• Reconstruct Surfaces that do not have support in the input 3d point cloud (low textured walls, windows)
![Page 4: CSE590 V : Multi-View Reconstruction](https://reader035.vdocuments.site/reader035/viewer/2022062410/568163e4550346895dd542fb/html5/thumbnails/4.jpg)
Robust and Efficient surface reconstruction from range-data[CGF- 09]
P. LabatutJ. P. Pons
R. Keriven
![Page 5: CSE590 V : Multi-View Reconstruction](https://reader035.vdocuments.site/reader035/viewer/2022062410/568163e4550346895dd542fb/html5/thumbnails/5.jpg)
Problem Definition
• Reconstruct a surface from a set of merged scans (noisy and outliers)
![Page 6: CSE590 V : Multi-View Reconstruction](https://reader035.vdocuments.site/reader035/viewer/2022062410/568163e4550346895dd542fb/html5/thumbnails/6.jpg)
Related Work• 2 Primary Themes– Implicit Surfaces (Poisson Surface Reconstruction)
– Delaunay Methods
![Page 7: CSE590 V : Multi-View Reconstruction](https://reader035.vdocuments.site/reader035/viewer/2022062410/568163e4550346895dd542fb/html5/thumbnails/7.jpg)
Algorithm
• Perform a Delaunay Triangulation/Tetrahedralization of the 3d point cloud + cameras/sensors.
![Page 8: CSE590 V : Multi-View Reconstruction](https://reader035.vdocuments.site/reader035/viewer/2022062410/568163e4550346895dd542fb/html5/thumbnails/8.jpg)
Algorithm
• Perform a Delaunay Triangulation/Tetrahedralization of the 3d point cloud + cameras/sensors.
![Page 9: CSE590 V : Multi-View Reconstruction](https://reader035.vdocuments.site/reader035/viewer/2022062410/568163e4550346895dd542fb/html5/thumbnails/9.jpg)
Algorithm
• Construct a directed graph– Each tetrahedron is a vertex
![Page 10: CSE590 V : Multi-View Reconstruction](https://reader035.vdocuments.site/reader035/viewer/2022062410/568163e4550346895dd542fb/html5/thumbnails/10.jpg)
Algorithm
• Construct a directed graph– Each tetrahedron is a vertex
![Page 11: CSE590 V : Multi-View Reconstruction](https://reader035.vdocuments.site/reader035/viewer/2022062410/568163e4550346895dd542fb/html5/thumbnails/11.jpg)
Algorithm
• Construct a directed graph– Each tetrahedron is a vertex
![Page 12: CSE590 V : Multi-View Reconstruction](https://reader035.vdocuments.site/reader035/viewer/2022062410/568163e4550346895dd542fb/html5/thumbnails/12.jpg)
Algorithm
• Find an s-t cutset of the directed graph– An additional s & t vertex
![Page 13: CSE590 V : Multi-View Reconstruction](https://reader035.vdocuments.site/reader035/viewer/2022062410/568163e4550346895dd542fb/html5/thumbnails/13.jpg)
Algorithm
• Find an s-t cutset of the directed graph– The s-t cutset gives the surface
![Page 14: CSE590 V : Multi-View Reconstruction](https://reader035.vdocuments.site/reader035/viewer/2022062410/568163e4550346895dd542fb/html5/thumbnails/14.jpg)
Formulation
: Visibility Information from points, cameras.
: Quality of reconstructed surface in terms of size of triangles.
![Page 15: CSE590 V : Multi-View Reconstruction](https://reader035.vdocuments.site/reader035/viewer/2022062410/568163e4550346895dd542fb/html5/thumbnails/15.jpg)
Surface Visibility
![Page 16: CSE590 V : Multi-View Reconstruction](https://reader035.vdocuments.site/reader035/viewer/2022062410/568163e4550346895dd542fb/html5/thumbnails/16.jpg)
Results
![Page 17: CSE590 V : Multi-View Reconstruction](https://reader035.vdocuments.site/reader035/viewer/2022062410/568163e4550346895dd542fb/html5/thumbnails/17.jpg)
Multi-View Reconstruction Preserving Weakly-Supported Surfaces[CVPR 2011]
Michael JancosekTomas Padjla
![Page 18: CSE590 V : Multi-View Reconstruction](https://reader035.vdocuments.site/reader035/viewer/2022062410/568163e4550346895dd542fb/html5/thumbnails/18.jpg)
Problem Definition
• Reconstruct Surfaces that do not have support in the input 3d point cloud (low textured walls, windows)
Input Image Point Cloud Reconstructed Surface using CFG 09
![Page 19: CSE590 V : Multi-View Reconstruction](https://reader035.vdocuments.site/reader035/viewer/2022062410/568163e4550346895dd542fb/html5/thumbnails/19.jpg)
Weakly Supported Surfaces
![Page 20: CSE590 V : Multi-View Reconstruction](https://reader035.vdocuments.site/reader035/viewer/2022062410/568163e4550346895dd542fb/html5/thumbnails/20.jpg)
Key Claim
• Large Jump in Free Space Support as we go from outside to inside.
• Even true for weakly supported surfaces.
![Page 21: CSE590 V : Multi-View Reconstruction](https://reader035.vdocuments.site/reader035/viewer/2022062410/568163e4550346895dd542fb/html5/thumbnails/21.jpg)
Weakly Supported Surfaces
• Ws = 9 ( 3 * 3)
![Page 22: CSE590 V : Multi-View Reconstruction](https://reader035.vdocuments.site/reader035/viewer/2022062410/568163e4550346895dd542fb/html5/thumbnails/22.jpg)
Weakly Supported Surfaces
• Wt = 3 ( 1 * 3)
![Page 23: CSE590 V : Multi-View Reconstruction](https://reader035.vdocuments.site/reader035/viewer/2022062410/568163e4550346895dd542fb/html5/thumbnails/23.jpg)
Weakly Supported Surfaces
• Ws > Wt -> creates wrong cut.
![Page 24: CSE590 V : Multi-View Reconstruction](https://reader035.vdocuments.site/reader035/viewer/2022062410/568163e4550346895dd542fb/html5/thumbnails/24.jpg)
Old t-weights
![Page 25: CSE590 V : Multi-View Reconstruction](https://reader035.vdocuments.site/reader035/viewer/2022062410/568163e4550346895dd542fb/html5/thumbnails/25.jpg)
Modified Weights
![Page 26: CSE590 V : Multi-View Reconstruction](https://reader035.vdocuments.site/reader035/viewer/2022062410/568163e4550346895dd542fb/html5/thumbnails/26.jpg)
PerformanceDataSet/Method Baseline[CFG 09](mins) Ours(mins)
Castle 30 32
Dragon 90 94
![Page 27: CSE590 V : Multi-View Reconstruction](https://reader035.vdocuments.site/reader035/viewer/2022062410/568163e4550346895dd542fb/html5/thumbnails/27.jpg)
Results
INPUT IMAGE POINT CLOUD
CFG-09 OUR METHOD
![Page 28: CSE590 V : Multi-View Reconstruction](https://reader035.vdocuments.site/reader035/viewer/2022062410/568163e4550346895dd542fb/html5/thumbnails/28.jpg)
Results
INPUT IMAGE POINT CLOUD
CFG-09 OUR METHOD
![Page 29: CSE590 V : Multi-View Reconstruction](https://reader035.vdocuments.site/reader035/viewer/2022062410/568163e4550346895dd542fb/html5/thumbnails/29.jpg)
Results
INPUT IMAGE POINT CLOUD
CGF-09 OUR METHOD
![Page 30: CSE590 V : Multi-View Reconstruction](https://reader035.vdocuments.site/reader035/viewer/2022062410/568163e4550346895dd542fb/html5/thumbnails/30.jpg)
Discussion
• How good is the claim about free space jumps ?
• What should be the multiplication term ?