Is there any general library or ROS package that is compatible with fisheye like distorted images? Until now I was undistorting my images and using the ar_track_alvar package but this sacrifices the wide field of view. And what intrinsic calibration information do such programs get? (I think OpenCV 2.x.x only supports pinhole models). An alternative to estimate a monocular fisheye camera pose would also be useful.
Originally posted by Mehdi. on ROS Answers with karma: 3339 on 2015-01-18
Post score: 1
Original comments
Comment by Falko on 2015-10-30:
Did you find any information about this topic? I'm also thinking about detecting markers in fisheye images. But I'm not sure if alvar can handle the distortion. Should I risk buying a fisheye camera and not being able to process the images?
Comment by machinekoder on 2019-06-26:
The topic is old, but you undistort the image with image_proc after calibrating the camera.
Comment by Mehdi. on 2019-06-27:
@machinekoder the whole point was about using the full field of view of the camera to cover more space. Undistorting the image will crop a lot of that field out view out. See my old answer.
Comment by machinekoder on 2019-06-28:
But keep in mind undistorting should not sacrifice FOV. It just undistorts the image of the fisheye lens, which itself already loses information compared to a less distorted lens (more details on the same image sensor area). So yes, the fisheye lens can indeed improve FOV compared to a standard lens, with the cost of image quality.