Tips for improving orthophoto from drone imagery

[orellius]

Hello,

I've been using Metashape about 6 years now. I fly multiple commercial drones, typically RTK/PPK capable systems with high accuracy positioning. I typically do surveys with it, so it's point clouds exported and deliverables extracted in Global Mapper.

Typically the quality of the orthophoto is not that important, so little oddities in the ortho are not an issue. Meaning objects/structures on site, like buildings and such don't have to look perfect, it's all about the accuracy of the elevations in the ground/terrain, nothing to do with structures.

I've done a few flights over complex sites, like small natural gas plants for example. I get all sorts of artifacts/oddities in the resulting orthophoto where all the details are, like pipe networks and cable trays and building/structures. I am just looking to improve on it. Hoping to get tips on how to process it to get a cleaner looking image.

Notes:

• Camera - Sony RX1 RII 42mp
• PPK positions on photos - position accuracy set to 0.015m in Metashape
• flown @ 61m/200ft with 80/70 overlap, with cross-grid so I have images perpendicular to standard grid as well. Generally I found this improves the results but still get some issues
• Metashape batch settings:
• Align photos - high, reference/generic pre-selection on, Key point limit 100k, tie point limit 40k, both those values are drone manufacturer recommendations, adaptive camera model  fitting enabled
• Optimize - everything selected except additional corrections, camera model fitting enabled
• Dense Cloud - medium, aggressive
• DEM - from dense cloud
• Ortho - from DEM, Mosaic blending mode, refine seamlines = no, ghosting filter = no
• I have experimented with the refine seamlines and ghosting filter enabled but still see similar results
• I have tried filtering the dense cloud to ground points only, then re-generate DEM and ortho from ground points, but the result is always worse, some detailed areas may look cleaner, but other areas there will be large splits in the ortho, like not missing info, just perhaps a vehicle is split in half and offset from the other half, so it's not a vehicle anymore, or a building split in half, hard to explain. Doesn't seem like filtering to ground points helps the DEM/Ortho generation much

I am not an expert with advance processing in the software, mostly over the years just using standard settings advised by drone manufacturers, automating the processing with batch files and just exporting the point cloud and ortho for volumes. So perhaps some of the settings above could be changed? Just looking for tips to create better orthophotos of detailed sites like this.

I feel like it is the DEM/point cloud causing the issues in the ortho image, but don't know how to fix it. I should have way more than enough overlapping images to create a clean looking result.

See attached screenshots, look at the pipe and building edges.

Thanks a bunch!
 

[orellius]

Further, is it my dense cloud and my computer's capabilities that are causing the issue? Like I usually generate medium dense cloud due to RAM/resources on my laptop. See another screenshot below of the point cloud, DEM, and resulting ortho on this particular dataset.

My current work laptop is a HP Zbook 17 G2 w/ Quadro K51000M graphics, 32 GB RAM. If I try processing drone flights on high dense cloud setting I typically don't have enough memory, or it will take way too long.

I have a new laptop already ordered, is coming in a few weeks. HP Zbook 17 G7 Fury w/ Quadro RTX5000 graphics, 64 GB RAM. Perhaps if I can create a denser cloud the resulting DEM/ortho will be better?

[Invaidor]

I use Sparse Cloud when i generate the DEM for the Ortophoto. genretates less (almost none) )artifacts in the photo.
My goal is alway to have 80% forvard fligt overlap and between 65-75% side overlap. at volymetric flights I fly crossgrid and at 5m/sek @ 120meter (150) height, and that will result in 90/90% overlap. But is werry unesseserry.

When I generate Dense Cloud i always use MILD filtering. too mutch of the details is removed for me.

If you give me an not to big set I can see what i can generate for you, and see what parameter to use.

[orellius]

I use Sparse Cloud when i generate the DEM for the Ortophoto. genretates less (almost none) )artifacts in the photo.
My goal is alway to have 80% forvard fligt overlap and between 65-75% side overlap. at volymetric flights I fly crossgrid and at 5m/sek @ 120meter (150) height, and that will result in 90/90% overlap. But is werry unesseserry.

When I generate Dense Cloud i always use MILD filtering. too mutch of the details is removed for me.

If you give me an not to big set I can see what i can generate for you, and see what parameter to use.

Thanks for the tips. I forgot about the depth filtering, I've had it set to aggressive for so many years due to multiple drone manufacturer's recommendations, but that's for typical surveys like gravel pits. The filtering is probably too aggressive and removing all the fine details of the gas plant site. I'm going to try re-processing it overnight with your suggestions and see what it looks like. I'll see the results tomorrow.