Hi,
I have a network of >10 synchronized cameras (with overlapping views)
and need to find the extrinsic calibration matrices. Since the we
need
to be independent of lighting, I like to use a LED wand which a user
waves in the perceived space. This yields the necessary point
correspondences.
After some research, I found several papers incor****ating this
technique (e.g. http://cmp.felk.cvut.cz/~svoboda/SelfCal/
or "Chen:
Wide Area Camera Calibration Using Virutal Calibration Objects") ,but
I am not sure which way to go, or what technique is more commonly
used
nowadays.
In my understaning, one way is to compute the Fundamental matrix for
each camera pair, and then
apply bundle adjustment to to obtain a "global" optimal solution for
the camera matrices, as well as the 3D points X.
Apparently, as an alternative one computes the camera matrices via
projective factorization in which projective depths are estimated.
Is one way favorible over the other. Which technique is more
computationally expensive?
Thank you for your help,
Tobias
|
