My name is Turcan, I am a senior PhD Candidate at Robotic Systems Lab, ETH Zurich. I am one of the leads in the project called GrandTour Dataset, where we have used one of your sensors (Zed2i Stereo USB camera) alongside others.
Given this context, as this was a dataset we have provided the “raw” images to the public as some users might want to rectify their images themselves or use the distorted images directly. Now the issue and the questions:
Despite my sincere efforts, I am unable to rectify a ZED2i provided raw left camera image to be an rectified image. I am using the camera-provided camera intrinsics parameters including the distortion parameters.
Conversely, if I look at the ZED-SDK rectified images they are just fine, infact, quite good. Implying to me that the camera is still well calibrated.
I have rectified other types of camera before hence, I would like to add that I believe I know what I am doing, yet, multiple people including me couldn’t rectify a raw image from the device.
Question & Request:
Is it possible for me to get more details on how StereoLabs rectify their images?
Was there any previous application where a user rectify the images themselves?
Is the Plumb Bob model, provided by the SDK / ROS drivers correct?
Is anywhere within StereoLabs software suite, there is an example how rectification can be done on your images? (E.g. with OpenCV or other tools)
The .svo2 files I am using has been collected using SDK v4.2 while I am using SDK v5.0.5 for deserializing the images from .svo2 file.
If helps, I can share one.
TLDR; ZED-SDK rectification works yet, manual raw to rectified image correction does not.