Forum

Show Posts

This section allows you to view all posts made by this member. Note that you can only see posts made in areas you currently have access to.


Messages - 4DS Niklas

Pages: [1]
1
General / Re: RTK Positioning accuracy with Mavic 3E
« on: January 30, 2024, 06:32:26 PM »
Also auch wenn du 10cm in den RTK Bildern hast passen die GCPs auf Anhieb?
Ich bin inzwischen soweit, dass ich die RTK Positionen am Ende ausschalte, und die Transformation nur noch auf die GCPs anpasse, weil ich sonst nicht so wirklich glücklich mit den Fehlern werde.
Den GCPs vertraue ich dann immernoch sehr viel mehr als den RTK Positionen der Bilder. Aber eigentlich ist das natürlich nicht Sinn der Sache.

2
General / Re: RTK Positioning accuracy with Mavic 3E
« on: January 30, 2024, 12:55:16 AM »
Hast du die automatische Verzerrungsanpassung im Kameramenü  aus?
Ja, die ist aus.

Hast du den mechanischen Verschluß eingeschaltet?
Ja.

Welche Verschlußzeit bei welcher Höhe benutzt du?
Normalerweise 1/500 bei 60m.

Wo hast du die Abweichungen, bei den GCP oder in den berechneten RTK-Koordinaten der Bilder?
In erster Linie wundert mich die Abweichungen bei den RTK Koordinaten der Bilder. Hier hatte ich bei der P4 sehr viel besser Ergebnisse beim ersten Alignment.
Die GCPs passen dann auch in erster Instanz entsprechend schlecht.

Wie ist die Überlappung in x und y?
80/80

Wenn ersteres, wie hast du die GCP bestimmt?
Normalerweise 1x 10 Beobachtungen mit der Emlid Reach RS3. Wenn es etwas genauer sein soll mache ich auch schonmal eine Runde vor dem Flug und eine danach.

Und als letztes: Machst du nur Senkrechtaufnahmen oder benutzt du Smart Oblique Fotos?
Die schlechter Qualität der RTK Positionen ist mir vorallem beim Smart Oblique Modus aufgefallen. Aber auch wenn ich nur Nadir schieße habe ich oft mittelmäßige Ergebnisse.

Hast du bezüglich des Verhältnisses Shutter/Fluggeschwindigkeit/Höhe Erfahrungen die du teilen kannst? Das würde mich besonders interessieren.
Danke dir
LG Niklas

3
General / RTK Positioning accuracy with Mavic 3E
« on: January 29, 2024, 05:51:24 PM »
Hey everyone,

until a few month ago, i used the Phantom 4 RTK for my aerial surveys.
The system estimated the images location accuracy to be around 1-2cm H / 2-3cm V.
When aligning images, i usually got RMS around 2cm which seemed plausible. The GCPs usually confirmed this.

Now, with the Mavic 3E RTK, i just cannot seem to reach this quality for the RTK data, even if the system itself estimates its own accuracy to be very good, sometimes even sub 1cm (lol, of course DJI)
But when i process the data in metashape, the ugly truth comes out: usually i get around 4-5 cm RMS, with outliers of 20cm errors or more.

Now with the P4, i used an SDK combo with the old GS Pro App, which let the drone stop for taking each image. The Mavic with DJI Pilot 2 just takes images while moving.
I assumed this might be a problem, but when i did some expirments, flying a small area with 4 good quality GCPs, with 3 m/s, 6 m/s, 9 m/s at 40m height above ground, there was no trend of increasing errors in RTK positions. Kinda frustratingly, during these tests, the RMS from alignment was very good, and GCPs also confirmed a pretty solid quality for the RTK positions.
But it was only a relatively small area, around 40-50 images. For larger areas, and especially when using the Mavics oblique capture mode, the RTK positions just seem to be all over the place.
I tried using Metashapes rolling shutter compensation, and it improved the results somewhat, but still nowhere near where they should be.

Does anyone else have any experiences with this? Another possibility would be ppm error from distance to the nearest reference station (I use the public german NTRIP service, Sapos).
But before i go off and doing a thousand tests to get to the bottom of this, i wanted to ask this very knowledgeble community if anyone had any ideas, or similar, or maybe even different experiences?
Please let me know.

Greetings from germany,
Niklas

4
Hey everyone,

i have had similar experiences with RTK drones. Especially the error in elevation between the tie points and GCPs pops up every now and again.

I am not a technical expert, but we assumed it has to do with the camera calibration parameters, especially f, cannot be estimated with 100% certainty with just RTK positions without the additional info GCPs provide to the bundle adjustment. I would be very happy to hear about this from someone with more technical knowledge.

I also find that the RTK Positions from the Mavic 3E are not of the same quality as the P4 RTK. When using the P4, the software we used stopped the drone for every image taken. The Mavic 3E with the preinstalled DJI Pilot app does not allow this. This makes the flight faster, but i feel the positioning accuracy suffers because of this. The Drone itsfelf always estimates its accuracy to be way higher than the survey results with GCPs show in the end. Did anyone else have this experience?
 

5
General / Batch Process Orthomosaics with planar projection
« on: April 13, 2023, 08:07:45 PM »
Hello,

i have captured a building with my drone and wanted to generate orthomosaics of all its facades. I transformed it to a local coordinate system so the building is parallel to XY.
Now i realized i do not have the options for front XY etc. when batch processing orthomosaics. Is there a solution for this?

I have found the following code in an old thread:

Code: [Select]
import PhotoScan

front = PhotoScan.Matrix([[1, 0, 0, 0],[0, 0, 1, 0],[0, -1, 0, 0],[0, 0, 0, 1]])
back = PhotoScan.Matrix([[-1, -0, -0, 0],[0, 0, 1, 0],[0, 1, 0, 0],[0, 0, 0, 1]])
left = PhotoScan.Matrix([[-0, -1, -0, 0],[0, 0, 1, 0],[-1, 0, 0, 0],[0, 0, 0, 1]])
right = PhotoScan.Matrix([[0, 1, 0, 0],[0, 0, 1, 0],[1, 0, 0, 0],[0, 0, 0, 1]])

doc = PhotoScan.app.document

for chunk in doc.chunks:

if chunk.label.endswith("East"):
proj = right
elif chunk.label.endswith("South"):
proj = front
elif chunk.label.endswith("West"):
proj = left
else:
proj = back

chunk.buildOrthomosaic(surface=PhotoScan.DataSource.ModelData,blending=PhotoScan.BlendingMode.MosaicBlending,color_correction=False,projection=proj)

Would this code still work, do i just have to replace photoscan with metashape? Would it actually do what i described above? Sadly, im not very well versed in python or how to utilize metashapes API.
Thanks in advance!
Niklas

6
General / Problems aligning big Dataset in one aligment
« on: August 03, 2022, 11:16:29 PM »
Hello everyone,

i have a dataset of ~2400 images capturing a schoolbuilding.

When trying to align these images, only 1750 got aligned in one attempt. I aligned the others in 2 seperate chunks, which worked fine.
I planned to try to align the chunks together using Align Chunks point based, but it also did not seem to work.

I have 2 chunks containing images capturing the buildings facades at low distances at mostly 0°, and one chunk capturing the building from above at -90° to roughly -50°.
In my experience, the overlapping areas should be more than enough to get the images to align in one go, but i cannot seem to get it working.
Any ideas what i could do differently to get these images to align in one go?
I attached images of the chunks where i marked the same facade in each for better orientantion.

7
General / Automatic detection of bad/blurry images
« on: May 24, 2022, 04:16:26 PM »
Hello everyone,

at the moment, i am working on a orthomosaic of a church's facade.
Sadly, some pictures seem to be out of focus.
I started to manually optimize my orthomosaic with the draw polygon + assign images tool, so far it seems there are better images available for most parts of the church, so hopefully i would not have to go get more images.
But i rembered, or at least thought i remembered, that i read about a  tool in metashape to automatically detect blurry pictures, which i could then disable for the ortho. Sadly, i cant find any more info on that.
Can anyone help me or was i imagining things?

Greetings,
Niklas

8
Hello everyone,

i have a giant set of 4000 images of a building of which i plan to generate a detailed pointcloud and a lot of orthomosaics of the buildings facades.
Because of vegetation the quality of the pictures location data varies a lot, with about 80% having precise coordinates and the rest varying between normal gps quality and no position at all.

In the past i had the problem in a similar case, where the program matched my images into two groups when i had source preselection activated.
I tried to get these to align together by manually creating markers, but it did not work at the time.
Any suggestions on how i would best proceed in this case? I also have around 80 points around the building measured with a total station, so i dont neccessarily need the images RTK data, but i assume using them would cut the time needed for alignment by a lot. Since its so many images, i wanted to try to get some input first, since its hard to expirement with this much data.
Thanks a lot in advance!

Niklas

9
General / Strange Noise in Point Cloud on new Desktop PC
« on: May 22, 2021, 05:49:18 PM »
Hello everyone!

For my latest Project, i need to document the facades of a church, specifically one of its 50m high towers. For that took around 900 pictures from different perspectives all around the tower.

I used control points in the form of markers which were already on the facade. I measured those in a local system using a Leica TS16 Totalstation. I got pretty good results for the transformation via those points (around 3mm error using about 10 points).

Now to my problem:
We recently transfered our metashape license to our new computer. It has a AMD Ryzen 9 5950X 16-Core Processor 3.40 GHz CPU, 128 GB Ram, and a Nvidia RTX 3090 GPU.
Now, on our old machine, we got a pretty good result. We processed the project again on our new pc, using only slightly different control points to change the coordinate system, but now, we have some weird noise around the edges of the tower which i can not really explain. Anyone got any idea what the problem might be?
Please tell me if you need any more Info! See the picture attached to see what i mean.


Thanks in advance!

Niklas

Pages: [1]