Graphics Reference
In-Depth Information
Image-basedApproach,” which systematized existing image-based rendering tech-
niques into something known as
image-based modeling and rendering
[Debevec
et al. 96]. The pipeline of image-based modeling and rendering is intuitive and ef-
ficient, so the technology quickly spread into the production of movies and music
videos. Debevec himself was excited to apply this technology in film making.
5.3.1 Origin of Image-Based Modeling and Rendering
Debevec said that his motivation for the 1996 paper was to create an animation
of flying around the clock tower at the University of California, Berkeley, where
he was studying at the time. Because of this, the paper introduces tools geared
toward creating an animation. Debevec included the animation in a short film
he directed called
The Campanile Movie
[Debevec 97] that was included in the
Electronic Theater at SIGGRAPH (
Campanile
is the name of the clock tower at
Berkeley). The following year, students who contributed to the creation of the
short film participated in applying image-based rendering to the production of
the movie
The Matrix
.
5.3.2 Pipeline of Image-Based Modeling and Rendering
The process of the rendering method presented in the 1996 paper by Debevec,
Taylor, and Malik consists of a three-stage process, or pipeline, described as fol-
lows.
Stage 1. Photogrammetric modeling: Recovering approximate 3D
models
The goal of this stage is to construct an approximate architectural model
of the scene from the captured photographs. This is not automated—it requires a
fair amount of trial and error by a human using specialized software. This soft-
ware assists the user in constructing the model and matching it to the images. The
model at this stage is assumed to be fairly simple, consisting of simple geometric
primitives. Primitives are parametrized polyhedral “blocks” arranged in a hierar-
chical structure maintained by the software. The parameters of the blocks, which
are values such as the height and width, etc., are variables that can be linked to-
gether and constrained in a way that facilitates matching the model to the captured
images.
The user starts by marking the edges in the captured images corresponding
to edges in the rough model, then creates the model and lays out the parame-
ters. Then the software does the work of stereo reconstruction, and projects the
reconstructed model onto the image. The user then makes any adjustments and
performs another reconstruction, and repeats the process as necessary. Figure 5.9
illustrates the process: part (a) shows the lines marked by the user, part (b) shows
