Revision history [back]

I'm in the same boat.

Ninghang said the existing model and checkerboard calibration can still be used as a good approximation if you mask off everything beyond 180° during the calibration process.

I'd like to try this method and see how it compares to a manually calibrated radial model. But before reinventing this wheel yet again... has anyone written a fisheye calibrater that allows masking off beyond 180 degrees?

I don't expect OpenCV's model to stay a close approximation past 180, but for my task I can live with that, it will also be interesting to see the divergence.

I'm in the same boat.

Ninghang said the existing model and checkerboard calibration can still be used as a good approximation if you mask off everything beyond 180° during the calibration process.

I'd like to try this method and see how it compares to a manually calibrated radial model. But before reinventing this wheel yet again... has anyone written a fisheye calibrater that allows masking off beyond 180 degrees?

I don't expect OpenCV's model to stay a close approximation past 180, but for my task I can live with that, it will also be interesting to see the divergence.

I'm in the same boat.

Ninghang said the existing model and checkerboard calibration can still be used as a good approximation if you mask off everything beyond 180° during the calibration process.

I'd like to try this method and see how it compares to a manually calibrated radial model. But before reinventing this wheel yet again... has anyone written a fisheye calibrater that allows masking off beyond 180 degrees?

I don't expect OpenCV's model to stay a close approximation past 180, but for my task I can live with that, it will also be interesting to see the divergence.

Edit: Ninghang's post was written in 2012, and OpenCV's fisheye camera model appears to have been added in 2014. 2014, so he can't have been referring to the current model. However, I was under the impression that OpenCVs fisheye calibration model is the one from same as this paper, which is "known to approximate a large range of fisheye lenses" but looks to me like it will start deviating from lens behaviour beyond 180°, so I assume TAXfromDK's question is still valid for OpenCVs fisheye calibration model, and Ninghang's answer might be a solution if you don't need acurracy beyond 180°. I could be wrong - I haven't tried this yet, I intend to once I find/write calibration software that allows masking.

I'm in the same boat.

Ninghang said the existing model and checkerboard calibration can still be used as a good approximation if you mask off everything beyond 180° during the calibration process.

I'd like to try this method and see how it compares to a manually calibrated radial model. But before reinventing this wheel yet again... has anyone written a fisheye calibrater that allows masking off beyond 180 degrees?

I don't expect OpenCV's model to stay a close approximation past 180, but for my task I can live with that, it will also be interesting to see the divergence.

Edit: Ninghang's post was written in 2012, and OpenCV's fisheye camera model appears to have been added in 2014, so he can't have been referring to the current model. However, I was under the impression that OpenCVs fisheye calibration model is the same as this paper, which is "known to approximate a large range of fisheye lenses" but looks to me like it will start deviating from lens behaviour beyond 180°, so I assume TAXfromDK's question is still valid for OpenCVs fisheye calibration model, and Ninghang's answer might be a solution if you don't need acurracy beyond 180°. I could be wrong - I I'm still learning and haven't tried this yet, I intend to once I find/write calibration software that allows masking.

I'm in the same boat.

Ninghang said the existing model and checkerboard calibration can still be used as a good approximation if you mask off everything beyond 180° during the calibration process.

I'd like to try this method and see how it compares to a manually calibrated radial model. But before reinventing this wheel yet again... has anyone written a fisheye calibrater that allows masking off beyond 180 degrees?

I don't expect OpenCV's model to stay a close approximation past 180, but for my task I can live with that, it will also be interesting to see the divergence.

Edit: Ninghang's post was written in 2012, and OpenCV's fisheye camera model appears to have been added in 2014, so he can't might not have been referring to the current model. However, I was under the impression that OpenCVs fisheye calibration model is the same as this paper, which is "known to approximate a large range of fisheye lenses" but looks to me like it will start deviating from lens behaviour beyond 180°, so I assume TAXfromDK's question is still valid for OpenCVs fisheye calibration model, and Ninghang's answer might be a solution if you don't need acurracy beyond 180°. I could be wrong - I'm still learning and haven't tried this yet, I intend to once I find/write calibration software that allows masking.

I'm in the same boat.

Ninghang said the existing model and checkerboard calibration can still be used as a good approximation if you mask off everything beyond 180° during the calibration process.

I'd like to try this method and see how it compares to a manually calibrated radial model. But before reinventing this wheel yet again... has anyone written a fisheye calibrater that allows masking off beyond 180 degrees?

I don't expect OpenCV's model to stay a close approximation past 180, but for my task I can live with that, it will also be interesting to see the divergence.

Edit: Ninghang's post was written in 2012, and OpenCV's fisheye camera model appears to have been added in 2014, so he might not have been referring to the current model. However, I was under the impression that OpenCVs fisheye calibration model is the same as this paper, which is (2007), "known to approximate a large range of fisheye lenses" but looks to me like it will start deviating from lens behaviour beyond 180°, so I assume TAXfromDK's question is still valid for OpenCVs fisheye calibration model, and Ninghang's answer might be a solution if you don't need acurracy beyond 180°. I could be wrong - I'm still learning and haven't tried this yet, I intend to once I find/write calibration software that allows masking.