Two homography to attain stereo normal case

[saching13]

Start with the why:

Current implementation uses single homography to align the right camera to the left camera to reach the stereo normal case by making the relative rotation between stereo camera's near to zero.

However this method highly depends on the accuracy of the relative rotation that we have estimated during calibration and error-prone when the stereo cameras are far from stereo normal case. To overcome this issue we can use two homography which rotates both the camera's images to attain stereo normal case.
Apart from the two homography we also try to undistort the image before applying two thomography.

The How:

To achieve this we use Bouguet's cv2::stereoRectify() or Hartley's algorithm cv2:: stereoRectifyUncalibrated() which uses 3D geometry and least square approach in estimating the required parameters to build two homographies to achieve the stereo normal case.

The what

Achieving stereo normal case makes the epipoles to move to infinity and helps in simplifying matching for triangulation and intrun helping us in obtaining better depth map.

[saching13]

@Luxonis-Brandon just adding you here

[Luxonis-Brandon]

Thanks. So actually I get auto-added to all new issues. Appreciate you looking out though. :-)

[Luxonis-Brandon]

This is implemented in PR #190

[Luxonis-Brandon]

Just as an update this has been merged to develop and is soaking there for eventual merge to main.