Is the rectifyStereoImages function or the "Stereo Image Rectification" simulink model more accurate?

Code:

https://www.mathworks.com/help/vision/ref/rectifystereoimages.html

Simulink model:

https://www.mathworks.com/help/visionhdl/ug/stereoscopic-rectification.html

I ran several sets of images that were taken with different lens angles, of 120 deg and 140 deg. The attached results were really different, however (baseline-anaglyph uses rectifystereoimages and simulink-anaglyph uses the simulink model)

Answer 1

[Tarunbir Gambhir]

The StereoImageRectificationHDL model is a designed to be compatible with FPGA-hardware and is focused mainly on low computation requirements. In the Inverse Geometric Transformation section of the example, measures have been taken to improve the speed of computation while compromising a little on the precision and accuracy, as compared to the MATLAB function "rectifyStereoImages".

At the end of the Simulink example, in Simulations and Results, the results have been compared to the output from the "rectifyStereoImages" function.

Based on these points, the "rectifyStereoImages" function should give you a more accurate result.

[Yuval Levental]

Thanks for your answer, it is really informative. Can you see this reply? I did use Simulink blocks to filter out the lines as seen in the attached image, but the rectification is still very off.

What Simulink blocks do you think I could possibly use to compensate for the significant offset between the left and right camera images?