Single View Modeling

Li Zhang, PhD candidate of Robotics

Source image:

This is The Winchester Thurston School on Morewood Ave, Pittsburgh.

Profile view:

Some textures:

VRML model: result.wrl

User Interface:

I use GLUT and OPENGL to design this user interface under Windows NT. It can zoom in and zoom out the source image so that user can specify points, lines and quadrilaterials. After that it will generate VRML model based on those information and the algorithms given in Single View Metrology, Antonio Criminisi, Ian Reid, and Andrew Zisserman, ICCV 99.
First, I choose x, y and z axis and compute the vanishing points Vx, Vy and Vz using Bob Collins' method.
Second, I specify one point on x and its projection in image. Based on this relation, I determine the scale factor a for Vx by solving Equation. Then, do the same jobs to get b and c for Vy and Vz respectively. Finally I get the 3x4 projection matrix.
Third, based on height information, I can compute the 3D position by Homology matrix. The height information can be obtained by cross ratio.
If I had more time, I would like to model a sketch or painting and add some non-quadrilaterial texture mapping. 

Li Zhang

Last modified: Sat Oct 23 01:42:31 EDT 1999