Lecture 8

Trees from images

first some background reading:

  • Reconstructing 3d tree models from instrumented photographs, 2001 Dorsey's lab. Have to know the camera pose, some manual input to extract the matte. Constructs a visual hull from the photographs then does a heuristic graph search to identify interesting points on the medial axis of the tree. Takes the medial axis as an axiom to the L-system, endows it ith buds an grows L-systems to fill in the rest of the tree.
  • volumetric reconstruction and interactive rendering of trees from photographs, 2004 Register pictures using modeling clay on sticks in the picture. Blend alpha mattes using a volumetric reconstruction approach. Build a model using billboards. Doesn't allow relighting.
  • Image-based plant modeling, 2006
  • Image-based tree modeling, 2007
  • Neubert 2007: Use and cite the image processing stuff. Volumetric reconstruction approach similar to 2004 paper, but take the opacities to create an attractor graph which is used in particle simulation to estimate the structure. Leaves added as texture mapped planes. User needs to sketch out tree, not tree and don't know regions for alpha matting. Does need some manual tweaking in spots.

Main Ideas

The main problems seem to be: how do you register the cameras? how do you extract the mattes? how do you reconstruct a model from the mattes? What kind of model are you going to reconstruct in the first place? What to do about twigs and leaves? All along one is concerned with how much user input is required and where to use a heuristic versus remaining precise.

alpha matting.
discrete volume rendering equation

for Neubert: particle systems, Runge-Kutta, optimization of linear systems, da Vinci's width relationshiop.

Unless otherwise stated, the content of this page is licensed under Creative Commons Attribution-ShareAlike 3.0 License