Hi everyone,

Here is a stupid question about the estimation of trifocal
tensor. I think this mail list should be the right place to get the
help.

I am writing a program to estimate the trifocal tensor using
the linear method mainly with ppp and lll configurations. I know this process
consists of three steps: normalizing features, solving a linear system of
equations and recover the original tritensor. I am sure the first two steps are
OK as I have test the result using mvl functions such as
TriTensor::image1_transfer(). But I've got some problems to recover the trifocal
tensor.

I use the equation (9.2) of Faugeras and Luong's
book(Geometry of Multiple Images, 2001) but can't get the right result. So my
questions are:

(1) If the transformations H,H',H'' in (9.2) are the same
to those used to normalize the image features? My idea
is no. As I normalize the line segments by normalizing its two endpoints, the
two transformations for points and line segments should be identical in one
image. But they are transformations for points, not lines.

(2) If the equation (9.2) is correct as I am a little
suspicious of its veracity.

(3) Please give me a short algorithm to recover the tritensor,
i.e. the third step.

Any help appreciated!

Leonard