CDA 6938 Final Project Triangulation from Point Cloud

1
CDA 6938 Final Project Triangulation from Point
Cloud

• Yiyan Xiong

2
Background

• Mixed reality need models matching the real scene
  for two reasons
• Solve occlusion problem by using matting
  algorithm
• Light the real object with the virtual light
  resource
• Rough models are
• enough for the
• requirement

3
Triangulation

• Step 1 Group scanned points into cubes
• Parallelism raw scanned points

4
Triangulation

• Step 2 Take mean value of points in one cube as
  a resampled point
• Parallelism cubes / resampled point

5
Triangulation

• Step 3 Generate faces
• Parallelism cubes / resampled point

6
Triangulation

• Step 4 Triangulation with the resampled points

7
Ray Tracing

• Algorithm
• Send out a ray for each pixel
• Find nearest intersection between the ray and the
  objects
• Compute color from the intersection point and
  light source
• Parallelism
• Pixel

8
Test Result

• Test scanned bunny with 40097 points

9
Test Result
Bunny with 428 triangles

• Bunny with 3630 triangles

10
Test Result

• GPU Triangulation

Tri. No. Group Mean Face Mesh (CPU)
276 1.480ms 0.381ms 0.442ms 0.008ms
428 1.489ms 0.387ms 0.651ms 0.013ms
928 1.392ms 0.425ms 0.863ms 0.025ms
3630 1.428ms 0.661ms 1.132ms 0.133ms
11
Test Result
Triangulation GPU VS CPU
12
Test Result
Rendering GPU VS CPU
13
Future Work

• Optimize the triangulation
• Merge the triangles with similar normal
• Speed up the rendering
• Bounding Box
• KD tree

14
Thank you!