Hi all,
i trying to understand the better workflow for a large photogrammetry aerea.
I working on area about 160000 square meter, i used ground control points to control about 11000 zenithal photos defined with mavic 3 E (20mp per shot). Til this point all works good.
I would like to add another 15000 45 and 30 degrees photos to define a large mesh with all the details.
I tryed to put all photos in one project but system crashed in alignement phase for out of memory problem - my ram is 128 gb.
Someone can help me to understand the best process forr these situations?
Thanks
Hi smartargets,
what are your alignment settings?
Memory consumption during the alignment phase depends on the number of images (not size) and the number of detected features per image. For the default setting of 40 000 features per image (Alignment setting), you can expect the following boundaries:
- 2,000 images - 16GB RAM
- 4,000 images - 32GB RAM
- 8,000 images - 64GB RAM
- 16,000 images - 128GB RAM
So, you can decrease the amount of features per images in alignment settings or you can try to align the another images in other project, export the registration, import the new registration to the original project and then align the components. As you are using georeferenced components, you can use Merge georeferenced components option.
Thank you Orthan,
i already tryed to align the images in other project then i exporte di component and import that component in the georeferenced and ground controlled project,then i tryed to merge component, is this the right process?
I noted that this process is very slow, i think that the only option is to decrease the amount of feature as you already told me.
No, I thought the option in the alignment settings. But for that both components need to be georeferenced:
Hi Orthan,
what i trying to understand if is it possible to define a component and used this one as master with geoirenferencied GCP and add to this “slave” components that works on the base of the original ground control points.
In my project i add about 3000 photos to a base component of about 11000 photos the results is tat RC generated a joined component but i lost all GCP constraints.
I hope to explain the problem in then right way…
Thanks a lot
It is possible. How you lost GCP constraints? Can you show here some print screen of it?
What were your steps and alignment settings? Was the “slave” component computed as Unknown for its prior pose?
How was the base component created? How were the GCPs measured?
In your case it should work like:
- Import first set of images and align them
- Import GCPs in correct coordinate system and place them correctly around the images
- Set the prior pose of the images to Unknown or set No for Use camera priors for georeferencing in the alignment settings and align the images again (it also depends how precise the image’s GPS is and you can decide, if you want to keep these positions on)
- Import second part of the images, set their prior pose to unknown and align (in this step you can also measure some of the GCPs over this part of the images, which will help in the alignment process)
- If the components are not aligned you can also used another control point to merge the components
You can consider also this helpful in your case: RealityCapture Help
in attached you can find two screens about GCP, till first step (zenithal shots) everythings works well, other screen is about step two when i added component abot 3000 photos.
In this case i use GCP measured by gnns using ground targets. I am confident to precisions of this step.
Base component was computed using georeferenced data of photos with GCP in one file.
Slave component was computed in other file but i left epsg 32633 wgs 84 but without GCP, i thought that i can redefine the precisione of this component in the master file (with master component).
honestly what i don’t understand is why the alignment process seems has impact on GCP position…i thought that GCP should work as constraints of the job…
What are your steps after adding new component? Are you doing the alignment again?
Also, you have placed the GCPs on small amount of images (some of them are placed only on two images) and I see you also didn’t disable the geoinformation for the images before alignment.
It depends on your settings and what weights are set for the images and GCPs. Ideally, it should work also with image’s geo information, but in mostly cases the precision of drone’s GPS is not so precise. But 200 m errors seems to be too big anyway. In which axis it the biggest error there?
Also, you have yellow and red triangles which means that the GCPs are not placed well.
Are you using RTK drone?
Hi Orthan, i put everything in prior pose unknown and moved GPC in local coordinates, things seem are going better, but the model is overturned the Z axis goes bottom.
Do you have idea whats the problem?
Thanks again
Hello,
the GCPs were originally in local coordinate system or in some global system?
The model is turned after alignment? Is this the same after another alignment?
How are the GCPs placed over the model?
i changed settings of the project to euclidean then i converted all gcp in euclidean with local coordinates, i done another alignment and the model turned after this alignement.
GCP are distributed all over the area.
Thanks
If you want to use local CS, then I suppose it will be better to start from the scratch. In this case there could be used some settings from first alignment in your project and they could casing such issues.
But it is also possible to rotate the ground plane, so after that it would be placed correctly. Then just remember to use correct export settings with correct output coordinate system.
With local coordinates I found a solution, I have two questions to.undestand better the process.
Gcp are constraints or is it better to define distance to “force” alignment and reconstruction dimension?
The component file that I exported has inside the photo database or I have to add photos when I import an external component?
Thanks
Because i exported a components from a file processed on other PC and software give me this message “import file decryption key is not valid”?
Gcp are constraints or is it better to define distance to “force” alignment and reconstruction dimension?
It depends what you need. Using GCPs you are getting georeferenced and also scaled model. Using just distances you are getting only scaled model. Also, with control points it will help in the alignment process.
The component file that I exported has inside the photo database or I have to add photos when I import an external component?
It includes the images (like the path to the images).
Because i exported a components from a file processed on other PC and software give me this message “import file decryption key is not valid”?
On the same machine are you able to import that component to a new project?
It does sound similar and could possibly be related to another issue where the following fixes helped.
FIX 1
Uninstall the older version of RealityCapture, and make sure that everything connected to that version is being removed. Those files from the previous versions
“Although I had uninstalled the older version of RC it seems like my PC still kept some old stuff and just installed the new version into that folder. Those files from the older versions seem to mess with the timestamps.”
Delete older RC-versions and re-install it again.
FIX 2
Remove the device from the Domain Controller and uninstall it and remove it from the registry, after new installation and online activation it may work.
The time may be a problem because it can be managed by the organization and I can be different from the official time in 1 sec approx.
FIX 3
Make it like a clean install (hold CTRL + Shift and run RC, then choose Make it like a clean install) and use Offline activation.
FIX 4
Check if RCNode.exe is blocked by system firewall. Then go to RC install folder and start RCNode - Windows show firewall notification and accept it for all networks. Additionally you can start RC as admin and in general settings do license change → application owner resign in, also charge for offline use sign in again.
FIX 5
This issue might be connected to some really old files located in the C:\Program Files\Capturing Reality
Please try to uninstall RealityCapture then head to C:\Program Files\Capturing Reality and delete everything you will find about RealityCapture also in ProgramData, AppData, and Windows registry. Then restart your PC and try to install RealityCapture again and run the same test.
FIX 6
Rebooting (and allowing Windows to apply an update) cleared this in my case.
These are some of the fixes which were reported by our users, and they helped them
Hi Orthan,
Is there a limit of number of images i can add to a project?
As i told i proceeding to add components to project derived from the sub-projects and i would like to understand which is the limit.
Thanks
Hi,
there are the mentioned limits for one component, but after the features detection, they should be already computed, so the limit will increase. But it is hard to say, what is the upper limit there, it depends on your machine and used settings.
my pc is AMD Ryzen 9 7950X 128 gb ram Nvidia Ge forse 4070Ti…
Hi Orthan, is there a way to transfer a reality capture project to other people, sort of archive or similar, in a way that other people could open the project and excrat model or orthophoto?
Thanks