I have massive problems aligning Drone flights shot using RTK. No matter the settings Reality Capture always makes a mess of this data and does not seem to use the accurate RTK data to its advantage.
This is the worst offender I have seen with Reality Capture.
Photoscan and Pix4D have no problems with this dataset.
Is Reality Capture simply not usable for Civil Engineering?
As this forum is completely broken and I can not post another question because the forum just gives me an unknown error I would like to ask another one here.
Does Reality Capture Only support Static ground Lidar scanners?
Been trying to get data to work with the DJI Zenmuse L1 mounted on a drone. It kind of works inside Reality Capture but then always fails at some point in the pipeline.
Hello @Muninn_Odins,
I would recommend checking out this article:
It describes what to do when the image distortion accumulates with the aerial data. The main take from it is to use ground control points, while disabling the geo-location from the cameras (changing the Absolute pose in the Selected inputs panel to unknown).
You can also try changing the prior hardness value that can be found in the Alignment Settings. It can be set to a really high value (e.g. 1 000 000).
Yes I followed this article, and it does not matter what settings I use it will always return unusable results. I have however not tried to put prior hardness to 1 million. Can you please explain to me what this will do as help files inside Reality Capture are mostly very unclear on what settings do and even more so what value is stronger for many settings. Like weight, is 1 higher than 10 or 1000 for GCP.
I have ground control points and very precise RTK. It just seems Reality Capture is not utilizing this RTK data. I have verified this dataset in Photoscan and Pix4D and they have absolutely no problems.
Adding ground control to this dataset made the same kind of mess so it had no solution to this dataset in Reality Capture.
Is it possible as well to fix my profile so I can at least post more then this one single post to the forums?
If GCPs are used, then it is recommended to change the Absolute pose of the cameras to unknown, otherwise, the positions will collide. But when GCPs are used, then the curvature should not show up.
About the weight (of any kind in calculations). Weight increases the importance of a value in the calculation. A higher weight means stronger impact (or we can say pressure since it’s weight). So when a value has a higher weight, it will be taken more seriously in the calculations.
About the posting issues. The unknown error shows up when you try to create a new topic? I am not able to look into it, but someone will after I let them know.
Yes Using GCPs and unknown for cameras it creates an absolute mess as well.
Are there any further suggestions, we are about to make decisions on purchasing multiple licenses of Photogrammetry software, but it has to actually work. This dataset has now been processed with Metashape and it returns excellent results but we want to be able to use Reality Capture as well.
Hi, have you also tried this workflow from mentioned article:
There are alternative steps, which you can use for a big long dataset:
Which drone are you using?
Yes we have tried all these and more and Reality Capture is never able to make anything usable.
It is the Matrice 300 and P1 camera.
Other datasets have worked ok, but they all work much better in Metashape especially this dataset, no problems calibrating the whole dataset at once.
How many images are used in that project? Do you have calibration parameters for that camera?
6500 photographs
When you tried alternative workflow, were also that parts bent?
We have managed to get 1-2 parts to work almost ok, but merging will go back to making a mess.
Again other datasets have worked fine, shot with this same drone / camera.
We have now as well tested datasets from fixed wing drone using Sony A7rIV camera and Reality Capture is making a mess of these datasets as well.
All of the above datasets have zero problems in Agisoft Metashape.
This is a problem of narrow long lines. If other datasets worked, is it possible to use calibration parameters from that projects into this line project.
The fixed wing project is also narrow long line?
Fixed wing we have tried both long flights and many back and forth flights. So you are saying Reality Capture does not support long lines at all? Will this issue ever be resolved or should we simply write off Reality capture as a tool to use?
Does Reality Capture not use the accurate RTK data embedded in the files?
No, I am not saying that. I just wrote that those datasets are problematic for photogrammetry generally as is wrote in the article kumate sent you.
We are working on improvement of this.
RealityCapture is using RTK data. It also depends, which settings are you using for alignment. Also, using precise calibration parameters could help here.
Any idea when these improvements will be? Any time soon or far into the future?
