Topics

1

General / Orthomosaic from large aerial dataset issues

« on: November 23, 2022, 12:58:38 PM »

Hi,

I'm having some issues with generating Orthomosaics from large aerial data sets.

Data is from a PhaseOne iXM 100 camera and flown top down and also oblique at 45 degree angles.
The dense mesh and the tiled mesh looks absolutely fantastic.
We also need to generate orthos from this data.

When I go the DEM route the estimated calulation time for one ortho is over 12 days! for around a 4000 image chunk.

So I decided to go the 2.5d Heightmap route and genrated a height mesh from the densecloud and used this as the base for generating the orthomosaic.
Timewise this works very well and the Ortho looks ok, apart from the edges of buildings etc.

I attached some screenshots

If you looks at the otho you can see the jagged edges at the roof edge. This is visible everywhere.
The mesh you see is the generated 2,5d mesh..  as you can see there you also have the jagged edges, so it's logical they are also visible in the otho itself.
Finally there's an image of the DEM which looks much smoother.

Using the DEM for the ortho results in long calculation times.

Does anyone know what the best route is to generate orthos from large aerial data? and getting better results when using the 2,5D mesh?

Thanks in advance.. 

2

General / Managing Large Datasets (with Phase One 100 Megapixel Imagery(

« on: November 14, 2022, 05:03:31 PM »

Hi Guys,

We are rather new to processing our datasets in Metashape.
The forum provides a lot of information for general workflows and challenges but we couldn't find any information regarding the processing of (really)large projects.

I would therefore like to ask if you can give us some advice on what to do and what not to do when processing large datasets with metashape.
Our data consists of mainly aerial data shot with drones. Our main camera for datacollection is the PhaseOne iXM100, so 100Mpx images.
All our data is flown with a DJI M300 RTK drone.

The average datasets we process are around 6000 images maximum (100Mpx), but our projects are getting bigger and although we do not mind the increase in processing time we do stumble upon some problems processing certain steps.

For instance, we aree currently gathering data to process a digital twin of a large area in the Netherlands.
The area is flown in 9 parts and the total number of 100Mpx images is 40.000.
We will also scan the entire area from the ground with a mobile mapping system.

The entire area is flown top down (so camera facing downward) and oblique at 45degree angle from every side like survey planes do.
To get a workflow going we are currently processing a dataset of a factory ground also shot with the PhaseOne iXM100 under the DJI M300RTK. Total number of images is 17.100
 
We stumbled upon a few challenges along the way but the entire alignment in the end all seem to go well. It’s the mesh/model generation step where we are currently having trouble.
After a day or 6 of processing the calculation is around 7/8th into the calculation the system just freezes every time.

As mentioned, the alignment went well and all images are calculated properly.
The challenge starts when we start to generate the Dense Cloud at medium (or high). The Depth Maps are created but halfway the generation of the Dense Cloud the system freezes (after six days).


We monitored the pc to see if we can find the issue. It seems to have more then enough disc space and does not run out of RAM.
Our specs are
- CPU: AMD Ryzen Threadripper 3970X 32-Core Processor    3.70 GHz
- RAM: 128 GB
- GPU: GFX RTX2080Ti

How can we handle these large projects properly? Is it even doable to process a project this size (and the upcoming one of 40.000 images) in one go? Or should we split the project in the end to make it more manageable?

I found the metashape scripts and see one to split a data set back into chunks. Is this the way to go? And if so, are there any recommendations on how to proceed and take into account? (overlap of the chunks, seamlines of the different chunks, can we generate the final outputs/deliverable as one)

The end product should be a Dense Cloud, Orthomosaic and tiled mesh (like cesium and tiled obj).

Also, since this is a rather large dataset I read that it’s better to use depth maps to generate the mesh so we do not generate a dense cloud in this case. Am I correct?

Thanks and looking forward to your insights!

Regards,

Remco