Changing stereo vision baseline via software

0
votes

I'm trying to estimate the distance with stereo system of far objects. I have rectified the images acquired after calibration process and I have built the disparity map.

Is it now possible to change the baseline (and the maximum depth achievable) through changing the stereo calibration parameters; for example vector T or image center Cx contained in camera matrix?

2
opencvcomputer-visiondisparity-mapping
Do you mean to ask if it is possible to cahnge the baseline? – jilles de wit
I want to estimate the distance of some objects in the scene. Baseline is an important parameter to do that: depth=(baseline*focal)/disparity. So fixed real baseline we have maximum depth range achievable. My question is: is possible to change via software this parameter to increase depth range? Is possible to to this changing stereo calibration parameters? thanks – Speed87
see my answer; in short: no. – jilles de wit

2 Answers

0
votes

Given that your hardware is unchanged, you cannot change any parameters related to that. So the baseline and camera matrices are fixed. Baseline, lens parameters and pixel size together determine the maximal depth resolution you can achieve.

You might be able to slightly improve your result through superresolution techniques or some other trick, but this will not dramatically improve your results. Especially for faraway objects.

0
votes

No, it is not possible.

By considering pinhole camera model and epipolar geometry:

  • WCS World Coordinate System
  • R Rotation of a Camera in WCS
  • C Camera Position in WCS
  • I Identity Matrix
  • K Camera Matrix
  • P Projection Matrix
  • B Baseline

Projection Matrix P = K R [I -C]

After stereo rectification:

The images have to be within the same plane (Coplaner images), and all the epipolar lines are parallel. Also, epipoles are at infinity.

What changes? What does not change?

  • R and K change for each camera.
  • B and C do not change.
  • P changes for each camera as well (P depends on R and K).
  • Both cameras get the same K and R.
  • P is different for each camera (C is different (Baseline))