Hi, I have a farm and consider to test out WebODM. It’s just for adding a new hobby during lock down and I don’t plan to make any money on this(the farm have minimal activity and consist mostly of non productive forest). The area is about 2,7km2 so a mapping will be several thousand images. I don’t have a drone jet but consider to buy a maveric 2 or maveric 2 air. I have a server with 2x Xeon E5-2620 and 128GB of ram. But I can add up to 384GB of ram in total. If the processing time is a day or a week is not important as long as it doesn’t fail a lot. There is several points in the area that is professional marked with high accuracy GPS that can be used as GCP. The goal is to have a better detail level than google maps. Is this a feasible project?
Sounds like a lovely farm.
I’d suggest looking into the Parrot Anafi as well.
The Parrot Anafi has quite limited availability here in Norway. Specs wise the DJI drones looks better or am I missing something?
All depends, haha. If you’ve satisfied with what you’ve selected, then you’re making the right choice.
Welcome! Yes, the specifications of this project sound very reasonable.
Would I need to buy more ram? From reading older post I think I may need it, but the split function may help me not needing it?
Happy new year everybody
I’m not sure if you would at default processing settings, but if you crank the feature-quality, pc-quality, and min-num-tiepoints you can certainly run yourself out of RAM…
You can always get more if/when you need later, right?
Yes, no problem adding more later. But if it would failed for certain I would have added up front. But I will wait and see then.
What will be your GSD?
I really haven’t spent much time going into these details yet so this is just guessing by now. If I understand GSD correctly I think 1 to 2cm/pixel would be great.
Thanks for answering my questions.
It is partially a function of sensor and flight height (at least on the “fine” end, you can always downsample to make it more coarse/generlized). Most flight planning apps provide a means to estimate this before collection
Consider an Autel Evo 2 - larger battery, stated flight time about 40 minutes, but in reality you will likely get 34 minutes of real flight.
Autel app has direct support for mapping missions - you do not need third party software like you would if using a Mavic Air 2.
For number of pictures: Serioiusly consider if you need accuracy of less than 1 cm/pixel. If you fly at the max allowed altitude (120 meters) you will cover a lot more ground with each picture compared to say 30 meters and having super high resolution.
Remember to set MAX Altitude in the app slightly higher than the intended flight altitude to avoid issues when running the mission.
Thanks for the Autel tips, it looks interesting. Unfortunately it doesn’t look like they have any distribution in Norway.
I have looked into apps for mission planning. The area i plan to map consist of fields going down to sea level and up to hills about 90m. I would expect it will be beneficial to keep the same height over ground at all time. So far I have just found this feature in DroneDeploy. Is this as important as I think?
I am not sure about the altitude. I live in Denmark, where we rarely have these high differences in height locally.
I would probably break up the flight into several missions - probably needed for battery reasons anyway - and keep each at the same average height over terrain. Since the drone uses barometric height estimation, it will not compensate for elevation when flying uphill.
You should then be able to gather all images into a single project an process them. Be prepared to assign sufficient disk space to the Docker container and lot’s of RAM.
For a start, I would fly a mission over a limited area at say 110 meters and review the results.
Also, as I just learned - be aware of the wind. I had to abort a couple of missions the last few days. Temperatures around 0 degrees C and wind gusts to about 13 m/s (46+ km/h) made the drone stop then drifting backwards during flight. I always try to start downwind, that makes it easy to retrieve the drone if it can not handle the wind. Never fly with the wind on the outbound leg, you may be forced to land in an unprepared spot if the drone starts burning battery fast going nowhere
regarding the chooses the drone. keep in mind that the autel evo and mavic air/pro have rolling shutter which is not the best for mapping.
better choose are Phantom 4 advanced/pro who has global shutter.
To add a little to this thread, I have a similar project, but am now at the point of doing the mapping. I’m using a Mavic 2P, and a crossgrid of my 30-acre (a dozen hectares) gives me something like 700 photos. I’m starting out with RAW, adjust them as a group, then convert to smaller JPEGS (“only” 29MB each!!).
The question I have is about how to split/merge these for optimum processing. My machine is running a Ryzen 7, 8 core processor, and I have 32GB of RAM. In my first attempt at this, I did a split of 100 images, with a 100 meter overlap. I ran it as with high-resolution mapping to include a DTM, downsized the images to 2048 wide, and it ran around 11 hours. The output was mostly OK, although one area ended up weird and I’ll need to do the run over, or add additional photos in that area.
Would I be better off to increase the split size? Downsize the images further, change the overlap, or what?
Any suggestions would be very helpful.
Well, a small update here… After running ODM for 6-12 hours on a few occasions, and then having the program either stall, or fail due to an out-of-memory error, I’ve decided to upgrade my machine a bit. I will install another 64 gb of DDR4 RAM, which (I sincerly hope) will give me a total of 96 gb, and should suffice until I get more than 1500 photos per batch. The RAM sticks are 2x 16gb and 2x 32gb, all the same manufacturer, speed, latency, and timing, as well as both being dual rank. In theory this should work, but we’ll see. I doubt it will speed up the processing much, but at least I should be avoiding the stalls and out-of-memory errors.
I’ll report back on the outcome.
A quick update… the additional 64gb of RAM was installed, though not without a few setbacks. On initial installation, the machine would not finish its initial boot, so I was left with a dead screen. In the end, I disabled RAM acceleration. All the RAM is rated to run at 3200, and does so when installed separately as 32gb or 64 gb. When all 4 slots are populated, it won’t run at that speed, so it’s back down to the default 2400 speed but with 96gb of RAM.
I ran my most recent set of photos, 728 in all, with “ultra” quality settings, and it took about 2 1/2 hours to finish with a nice orthophoto. I’m also doing a shape-file of 2-meter contours, but who knows how long that is going to take.
Sounds like Ryzen, alright Giving me flashbacks to working in retail IT and troubleshooting everyone’s custom builds.
At the very least, I doubt you’ll ever feel the 800mhz difference in RAM speed in any real workload, so hooray, you’ve got 96gb!
Have you ensured that firmware is up-to-date and all modules were on the certified list for your motherboard? If so, then that’s just really bad luck and they shouldn’t be misbehaving like that.
Glad you’re up and processing, however!
You’re likely right about the memory speed issue. The fact that it was able to process 728 Jpegs that ran close to 14MB each is impressive by itself. But I was quite surprised that it could do it in just 2 1/2 hours. It did take another 30 minutes, give or take to generate the contour isopleths, but I consider that to be reasonable at this point. Perhaps one of these days I’ll tweak the speeds, but for now I’m satisfied.
Next I need to learn how to use QGIS for organizing the map and other details that I want to add about the property. Fun times!