I am filling a form about joining a student team. There are some questions about computer vision. I had a little experience before hand, and i spent two days already researching the basics about depth estimation. I have the following question about the implementation of depth estimation.
Lets say i have a moving vehicle, and i am using cameras to estimate depth of various objects. Knowing the fundamental matrix, depth estimation boils down to solving a 3x3 linear system for each point in space. (Assuming there are no uncertainties with the pixel coordinates, or distortion from lenses).
The fundamental matrix itself is calculated using corresponding points from the 2 images. The question is, do i have to calculate the fundamental matrix at every moment (every iteration)? Do i calibrate the cameras at the initial setup and after that, the fundamental matrix is considered known?
Once you calibrate your system, if extrinsic (relative position) and intrinsic (camera lenses, focal length, image resolution) parameters remain the same, then you don't need to recalculate the fundamental matrix F. You do it only once, and may store it for future use.
This is exactly what I do in my library, where you can find also an example where the fundamental matrix is calculated and used to draw epipolar lines.
I hope this answers your question.
Cheers