Camera Calibration from Coded Targets only?

[psapirst]

Is there some method in PhotoScan Pro to calculate the camera alignments from the array of coded targets?

As far as I can tell, the program is able to recognize the targets automatically, but will only compute the camera alignments using the normal target-less alignment procedure by matching common points among the photographs. However, this must decrease the accuracy of the alignment: while the target centers can be positioned very precisely at sub-pixel accuracy, the common points are much less precisely located. In practice, when working with a network of targets, I am finding measurement accuracy only in the 1:5,000–1:10,000 range (relative to the camera sensor). However, according to the published literature, calibrations using coded targets alone should allow measurement accuracy up to about 1:50,000 with a consumer dSLR camera.

Adding the ability to calibrate and align the cameras in the scene first by using the coded targets only would thus be very useful. With such an option, there would need to be an intermediate step to generate the sparse point cloud from the normal alignment step (but without changing the camera orientations).

[Porly]

+1

it would be helpful to have the possibility to align photos only with coded targets and let the natural feature extraction / matching only for the dense image matching process. Not only for camera calibration. Center estimating of coded targets is more exact (1/50 px) than the matching of natural features. Would be nice for close range photogrametry.

[Alexey Pasumansky]

Hello psapirst,

Currently it is not possible to use only markers for the camera alignment, but you can optimize the alignment results based on markers with very high projection accuracy parameter used.

[Porly]

Hello Alexey,

Hello psapirst,

Currently it is not possible to use only markers for the camera alignment, but you can optimize the alignment results based on markers with very high projection accuracy parameter used.

What would be the recommended projection accuracy in case of coded targets? Better than default 0.1px I guess??

[Alexey Pasumansky]

Hello Porly,

Yes, this parameter may be adjusted to higher accuracy, however, I don't think it is reasonable to go higher than 0.01, as it seems to be maximum that could be reached for coded target center detection accuracy in ideal conditions (moreover, the tie point accuracy default values are already slightly lower than common values).