Problem Statement

1
Problem Statement
Future Work

• Automatically adjust texture for lighting
  conditions
• Have algorithm interpolate new points to make a
  dense wireframe model
• Expand input set to arbitrary videos or pictures

• Construct a 3-Dimensional Model of a subjects
  face from 2-D images or video
• Could be useful for face recognition, e.g.
  identifying suspects from surveillance footage

4) Output

• Apply texture to model
• Render using OpenGL
• Able to interpolate new views

Algorithm

• Assumptions
• Cameras may be uncalibrated
• Certain views should be available
• Favorable lighting conditions
• No occlusions

3) Model Creation
1) User Input
2) Estimation

• Automatically estimates 3D Coordinates using a
  pair of images
• Based on algorithm proposed by Weng, Huang, and
  Ahuja, 1989
• Estimates camera parameters to within a scaling
  factor
• Merge 3 estimates by minimizing mean square error
  to get robust model

• User loads various poses of subject
• User marks several important points on each view

Special Thanks to Dr. Tsuhan Chen, Todd
Stephenson