Best practice to create a huge (12 km2 / 3000 acre) ortho-image?

Hello y’all! :slight_smile:

This year we have started commercial ortho-imaging services and did our first job for our client in that field. Our first order was quite big - to scan and create a geo-referenced orthophoto from about 12 km2 / 3000 acre area for our local municipality.

We performed the project within two long (10-12 hr) days in May & June in the endless light of Finnish summer, here near the Polar Circle, with one Mavic 2 Pro and about 15 batteries per day. We were flying from 150 m AGL, EVLOS, with all the necessary permissions from local Air Navigation Services.

I thought that flying the whole area in a one map is a no-no (I was afraid of that something goes wrong and we need to re-fly the whole thing), so I split the are to 6 sub-maps (about 2 km2 / 500 acre each) with some overlapping around the borders of each map.

Our client provided us 16 GCPs for the whole area, but unfortunately only 3 of them fell under each sub-map.

DroneDeploy did process the georeferencing with just 3 GCPs, despite the guide telling to have a minimum of 4.

All seemed to go good, but now in the autumn, when our client started to use the maps, they found that the georeferencing is way off - something like 5 meters in the edges of the maps, rendering the ortho-images unusable in their GIS application.

Another problem they were not happy about, was that the transitions between each ortho-image from the sub-maps we’ve shot individually, were quite too visible, as there were (of course) different lighting scenario when shooting each map, due to different time of day when shooting them.

Our client would want to have the borders of each ortho-image “mosaiqued” so that the borders would not be straight lines, but instead maybe softer ones, or following some terrain details like roads or building edges, thus making them less distracting when viewing the whole 12 km2 area.

I have just asked tips from DD support what I could do at this point, but I thought to share this issue also here in the forum, to hear if you guys have suggestions how these kind of huge (3000+ acre) areas would be best captured and georeferenced – in future projects, should I shoot one huge map with several thousand photos, or is there some way to combine smaller maps together to a more seamless, big, (accurately) georeferenced OrthoTIFF?

Thank you a lot in advance for your thoughts! :slight_smile:

Happy autumn,
Sakari, Finland

1 Like

Hi @Talviomena, welcome and happy Autumn! Very much needed here in Central Texas. I am jealous of your 150m AGL flight restrictions. Very large maps would be so much easier if we could fly that high.

Good move. This makes the whole mission more manageable and I am not sure what the shape of the subject is, but I don’t know how you would do a continuous mission over that large of an area let alone over multiple days.

I think this is where it fell apart. You definitely want at least 4 GCP’s encompassing the subject area as best as possible. Also, when stitching individual maps together you need to have GCP’s in common with adjacent maps which means you need to over-fly each map to get at least 1 row in common and 2 GCP’s.

Horizontally, vertically? The amount of error is typical in a non-GCP flight so this becomes strange.

My first reaction is plan better. Instead of trying to fly the project as quickly as possible try planning for the more consistent conditions. If you stretch an entire day you are going to get obviously different sun positioning in at least 3 phases. Stretch that across two days and you map will look like a quilt. Instead try mapping from 9:30am to 1:30pm or whatever you solar noon is and don’t be afraid to do it over several days of like conditions. Then…
You might consider a photo editing platform. Actually you NEED to get a photo editing platform… We use a software named ACDSee which is very affordable and offers perpetual licensing if you prefer. It allows you to perform non-destructive (maintains an original) batch editing in which you can setup standardized filters which will normalize the imagery. It breezes through larges projects while you do something more productive.

How many images was the project in total? If you would like to PM me I would be happy to run your imagery through the process to see if we can salvage what you already have.

I think the way you did it is fine. If your client has a GIS platform they can manage it from there. When we have projects of this nature and the clients do not have a GIS solution we create a master map in QGIS for them. Of course these files can become insanely large quickly so typically it ends up being a multi-page PDF with the master map being a much lower resolution key map.

Congratulations on the new venture and happy flying!

Thanks @MichaelL for a comprehensive and quick answer! :slight_smile:

This altitude limit is good indeed! Unfortunately we cannot enjoy it long more, as EU regulations will soon override our local regulations, and the limit will probably fall to 120 m.

Good to hear I made a right choice :slight_smile:

Ok! We might still have the possibility to get more GCPs measured by our client who has the precision equipment needed, and re-reference the maps. I luckily overflied each map so we might find common GCPs for both maps at least in some areas (there were also some areas where the map border / overlapping areas for adjacent maps were inside a vast forest, so there we might be out of luck finding visually distinguishable GCPs). But very good to know - in the future, we’ll definitely need to plan the maps in a such way that we can insert enough GCPs and also have these common GCPs in the overlapping area.

Horizontally. It might be maybe that the GCP calculation did not work at all in DD as I did not follow the rules to have at least 4 points, despite DD seemingly did something with the 3 GCPs

I agree! We definitely have to focus on that in future projects. In this case we actually had very limited options, as the area we scanned is just besides the busiest airport of Finland, and just under the glideslope for the two main runways. Normally we could not have ortho-imaged here at all (over 50 m AGL), but due to coronavirus and drastic reduction in traffic, we got an permit of exception to fly the mission there in 150 m, but only in those days, when the wind direction allowed the flights to be routed to another runway and with direct communication from us to the ATC tower during flight. There were very little options for such days (when the weather would also be suitable for drone operation), so in the uncertainity on which day we would be allowed to continue again, we needed to push as long days as possible.

Thanks for the tip! :slight_smile: We have Adobe CC package for other work, i.e. Photoshop & Lightroom, so these might work :thinking: though I have to study how we could normalize each OrthoTIFF. They are huge - 3-5 GB each, as they are with 5 cm / px resolution. But generally I feel they are quite near each others considering the overall color / exposure.

So, maybe with some kind of smoothing just the transition area between maps, maybe by mosaicing the edges, might do the trick?

Do you know, is this fine-tuning of the map edges possible to do with QGIS to our OrthoTIFF files and re-save them afterwards?

Thank you a lot for your generous offer, I appreciate it a lot! :slightly_smiling_face:
Unfortunately there are 6328 photos which take up tens of gigabytes, so it would be a relatively big process. Also the resulting OrthoTIFFs are multiple gigapixels & tens of gigabytes in total. But if we cannot solve this in our end, I’ll PM you and we can discuss more about possible options :slight_smile:

Good to hear :slight_smile: Our client uses Bentley MicroStation as their GIS application, and probably they could technically fix these, but as the order for our company was to provide ready-to-use OrthoTIFFs, we want to save their time and fix them ourselves as far as possible :slight_smile:

But as far as I’m concerned, the biggest issue is the horizontal error in georeferencing which prevents the usage of the images completely. The transitions between maps are luckily a cosmetical thing - nevertheless important also.

By the way, could it be possible to re-georeference the OrthoTIFFs we already have, with QGIS, if we get more points to use? :thinking:

Thank you very much! :blush: All the best to you and stay safe in these abnormal times!

Lightroom would be a great option. The only problem I had with it and the package is the move to a subscription base which I don’t really need since we use Bluebeam and if a photo editor works it works.

I correct the photos before I upload them so when Orthomosaics come out they’re done.

Maybe just select one mission that was the worst in uniformity…

Yes, you can geo-reference the images again in QGIS. Especially if you can see the GCP’s and have the coordinates for them that you need.

Ok, thanks again for the answers! :slight_smile:

I’m sorry I was a bit unclear earlier - all the missions are great in uniformity and there are no visible problems inside any resulting orthophoto. The challenge is to fine-tune the transition between adjacent orthophotos that are scanned on different days (and even on a different month)

This is great news! I could try that… Thanks again! :slight_smile:

1 Like

Ok so an alignment issue and not a photo exposure issue? If you used GCP’s that shouldn’t happen, but if there weren’t enough then I fear that the orthos may be skewed so even re-referencing them may not fix it.

Really, really silly question perhaps, but is it worth trying to process again but without the GCP input and align them manually? Just to see if you still get the same massive changes in size between maps?

I don’t generally use GCPs, I quite often don’t have the facility to, but when processing a same site multiple times and at different times of the year, quite large areas too, never have I been faced with a stretched map to that extent. Maybe half a metre perhaps? It might be worth forgoing the GCPs if there aren’t enough of them, just to see if it works.

I would say you are really lucky if you’re only experiencing less than a meter error between maps. The general accuracy of a drone is 1-3 meters and once those images are processed to a full map the entire map can shift up to 3m so a 1m error is pretty exceptional from map to map. They do have the new auto map alignment, but in my experience it is not very good especially on large maps. I think the intention of this mission was to use GCP’s for the needs of the client so I don’t think taking all the time to reprocess that big of a dataset is really going to solve any issues when it is obviously the improper use of GCP’s that cause the issue . I do agree with your suggestion to just see the difference in the processing, just not in this case.

The maps themselves may well be out of sync within DroneDeploy when flicking between one map and another online, but when exporting and importing into CAD for example, they get moved about in Model Space and overlay well.

The dimension between two identical points at the outer extents of two different maps processed on different days without GCPs might be up to half a metre. I have always been impressed by this.

1 Like

I’ll have to get your workflow sometimes because in our experience bringing in drone data to a state plane coordinate system is never just an overlay . Have you tried QGIS? I only ask that because using QGIS you go the other direction . You natively bring in the geotiffs and then bring in the CAD linework over top of that. It’s not part of our normal workflow, but it is another option. As I stated above you can also re-georeference the orthos in QGIS so if you have GCP’s in particular you can put it right on that basis and maybe it will line up better, but the issue with vertical is another matter.