Description
The PC_Demo is designed to perform our implicit surface reconstruction algorithm for unorganized point clouds based on convexified image segmentation. It is not optimized for performance. In fact, it was developed as a research code for the purpose of testing and experimenting with the ideas presented in the reference [1]. Our method can handle data sets that are non-uniform, and with holes or with open boundaries.
Algorithm
- Initial Processing
- simplify data point, average neighboring normals if necessary
- compute distance function and closest point information
- compute anisotropic Gaussian function for each data point
- compute the inner product field as the initial image
- Binary Image Segmentation Algorithm using TVG-L1 model (twice)
- Isosurface Extraction, the level set with value 0.5 is extracted
Experiments
Here are some experiment figures and a table of computation time (in seconds). All experiments are performed on a laptop with an Intel(R) Core(TM) i5-2430 CPU @2.40 GHz processor and 8.00 GB RAM memory.
|
point size |
point used |
grid size |
initial process |
TVG-L1 |
total |
dragon |
1,769,269 |
170,188 |
257x181x117 |
8.71 |
3.74 |
12.45 |
Bunny |
362.271 |
157,860 |
257x255x201 |
12.82 |
21.63 |
34.45 |
Bunny (holes) |
341,387 |
149,954 |
257x255x201 |
12.47 |
16.74 |
29.21 |
knot |
9,785 |
8,174 |
127x129x63 |
2.01 |
0.92 |
2.93 |
Venus |
44,992 |
43,820 |
181x251x257 |
16.16 |
6.77 |
22.93 |
Armadillo (non-uniform) |
1,142,120 |
111.377 |
217x257x199 |
11.68 |
13.21 |
24.89 |
Armadillo (sparse) |
22,403 |
20,456 |
217x257x195 |
10.83 |
13.64 |
24.47 |
- open surface (data with open boundaries)
- non-uniform and sparse Armadillo
Reference
Contact us
Jian Liang
Frederick Park
Hongkai Zhao
Acknowledgments
We would like to thank Edward Castillo for graciously providing us with his code for computing consistent surface normals to PC data.
|
|