Good XY data, struggling with Z.

[Unknownid]

Hey guys,

So im trying out agisoft for the company and have run a few different flights. My process has essentially been this http://www.agisoft.com/forum/index.php?topic=4229.0

I use RTK for my gcps (about 16 on the site 1kmx200m roughly) very flat cleared land (see report). The ortho comes out very accurate but based on the topo our field guys have done we're seeing discrepancy in parts of the dem in the z values. (~10cm).

Is there any  suggestions to fix this? In some parts (where theres lots of coverage) the accuracy was around 1cm. Discrepancy is closest to the non over lapped area.

thanks

[Paulo]

I would suggest to build Dense Cloud with high or medium quality settings instead of Low. This should improve Z accuracy....

Also the hero3 + is I think a Fisheye camera... In Camera calibration use Fisheye model instead of Frame thisi can improve results...

Regards,

[Unknownid]

thank you i will try this

[jmgc]

Any news on this subject?
I'm also dealing with DSM deviations.

I've been testing similar situations and Photoscan presents considerable errors on Z coordinate.
I've used GCP collected from a reference orthophoto and then used this GCP with another datasets (from the same location) hopping to get XYZ near perfect matches.
In XY I've noticed deviations of 30-50cm, in Z I've found 1m or more.
Any guess to this kind of problem?

[stihl]

Without knowing how many GCP's you used, how they're distributed and very importantly with what accuracy they were measured we can't answer any of your questions really..

[jmgc]

Without knowing how many GCP's you used, how they're distributed and very importantly with what accuracy they were measured we can't answer any of your questions really..

I've used 10 GCP, well distributed along the dataset, 200 to 300m from each other.
The reference ortophoto have 25cm pixel, I'll suppose the accuracy would be at least 25 cm.

[stihl]

In the best case it can't be better than 25 cm.
What's the GSD (resolution) of the project? The GCP distribution seems good.

How update is the reference photo? 25 cm pixel is Google Earth material. Maybe it's outdated from reality?

[jmgc]

In the best case it can't be better than 25 cm.
What's the GSD (resolution) of the project? The GCP distribution seems good.

How update is the reference photo? 25 cm pixel is Google Earth material. Maybe it's outdated from reality?

GSD (say ground resolution?) is 5cm.
The reference is an orthophoto from the same location flown a month earlier (processed under same conditions).
The GCP was extracted from this orthophoto.

[stihl]

Hmmr I'd put the marker accuracy at 0.001 m then and optimize it on that.

If the accuracy of the gcps is 25cm you can expect your output to have an accuracy 1-2 times this number for X and Y and 2-3 times for the height.

[jmgc]

Hmmr I'd put the marker accuracy at 0.001 m then and optimize it on that.

If the accuracy of the gcps is 25cm you can expect your output to have an accuracy 1-2 times this number for X and Y and 2-3 times for the height.

Thanks for the inputs stihl.
I've found Z differences of 1 m or more at points located over roads (unchangeable object).
Even with the 25cm accuracy I think it shouldn't be happening.

[dcm39]

Are all the images acquired at nadir or near nadir? With such a flat region, I would expect significant tradeoff between the radial distortion and focal length (http://doi.wiley.com/10.1002/esp.3609). Adding some significantly oblique images could help to resolve this. The "optimize" approach is a good after the fact solution, but I suspect that tough projects will break it.

[jmgc]

Are all the images acquired at nadir or near nadir? With such a flat region, I would expect significant tradeoff between the radial distortion and focal length (http://doi.wiley.com/10.1002/esp.3609). Adding some significantly oblique images could help to resolve this. The "optimize" approach is a good after the fact solution, but I suspect that tough projects will break it.

dcm39, the photos are acquired at nadir/near nadir. The only oblique photos are those caused by windy conditions.
I've found that the error is lower in areas with more changeable topography, which goes in line with your opinion.
Still testing and trying to get specific feedback from Agisoft support.
Thanks for the inputs.

[dcm39]

Yes, quite right. Depth of field is necessary in the images to break the tradeoff between radial distortion and focal length, so more topography will have a  similar effect to adding a number of oblique images.

[Bing66]

HI Just posted a similar problem, x and 7 errors below 2cm, z = 1.1 m, this is for a client, and its a stretch of road, but only photographed from 1 side, as the other side was inaccessible