Problem of Cropped FOV When Enabling Undistortion on OAK-1-W-97

qian

Hi,

I am using the OAK-1-W-97 camera, and I have an issue when enabling image undistortion. What I want is to turn on undistortion while keeping the full original FOV without any part of the image being cropped.

I followed the official guide and set i_resize_mode to LETTERBOX, which is supposed to preserve the full field of view. However, at the native resolution 1280×800, this setting has no visible effect. There are no black borders added, and the FOV is still cropped after undistortion is enabled. I’ve already verified the parameter is applied correctly with ROS 2 commands and restarted the driver, but the image behavior stays the same.

When I test other non-native resolutions such as 1280×500 and 1280×400, black bars do appear on the frame. Even so, the actual FOV is still smaller compared to the raw image without undistortion, especially the vertical view is still being cut off.

I would like to ask if there is another configuration or parameter that allows undistortion to run without cropping the image. I don’t mind having black borders around the frame, as long as the complete original FOV can be retained. Could you please advise if this is currently supported on OAK-1-W-97, or if there are any known limitations?

OskarSonc

Hey qian - if you are using the v3 depthai_ros_driver path, what you are seeing is expected right now.

i_resize_mode=LETTERBOX only controls how the final image is resized into the requested output size. It does not change the undistortion projection to “keep all original pixels”. Because of that, at native 1280x800 there is nothing to pad, so you won’t see black borders, and some FOV loss can still remain after undistortion.

For smaller outputs like 1280x500 or 1280x400, the black bars come from the resize step only. They do not restore the FOV already lost during undistortion, so the vertical view can still be smaller than the raw distorted image.

If full FOV retention is required, the practical workaround is to publish the distorted image and undistort on the host side with OpenCV using an optimal new camera matrix (for example with alpha=1), which will preserve the full image at the cost of black borders.

Thanks,
Oskar

qian

Thank you very much for your answer

OskarSonc

qian - happy to help!