We need to survey about 23,000+ Hectares of agriculture land and wondering what drone options we got given
(a) We have to survey really large area and small amount of time.
(b) We need to scan whole area say 3-4 times in a 8 months, over the season of crop to record growth.
(c) RGB + NDVI camera is good enough for now.
(d) Drone which is easy to repair, as I am sure there will be many crashes during operations..
Would it make more sense for to assemble it our own given we would need bunch of them, and we would need quick turn around time in case of repair/crash...
Replies
From AgPixel FB page:
@Darius, Your comment, "Large project like this, don't get awarded by gov agencies to hobby modellers
like @RC or us."
Very interesting observation, but so true. However the Easter Island survey would seem to refute the notion that 'hobby models' can't do such a project. It's on the scale of 20k ha and was performed in 1 week by two fixed-wing drones.
Maybe this is why governments around the world are bankrupt - they can't seem to overcome their own biases that lead them to spend money in such a wasteful way, they ignore capitalistic solutions in favor of bloated, corrupted institutions that they have always used in the past.
you are exactly right.
Large project jak this, don't get awarded by gov agencies to hobby modellers
like @RC or us.
Large projects get awarded to large contractors, universities, R&D institutes under public tender procedure.
So this challenge by @RC is all about it, how to Make Idiot Busy, to generate traffic.
23000 hectares or 10 km x 23 km area is large area by size, by access problems, by logistics.
100-man team can be busy for months to accomplish such task.
UN Agencies, World Bank manage projects of this size, attracting hundreds of local volunteers, students, R&D workers from universities and it takes months and MM$ budget spent.
Ok, you need 100 small drones and 200-man staff and M$5-M$10 budget to ink that contract and set deadlines.
You are exactly right, you can start as 3-man team with 5 drones
to test yourself.
But money, budget, funding is not a problem with large R&D projects, by public agencies.
23 km x 10 km area is large, very large by its size, by logistic and operation research standards.
UN Agencies, UN Foundation, World Bank or Google Earth Team, OpenMap Earth, NASA, NOAA, EPA
are the right contact places
@Darius, I only have experience with AgiSoft PhotoScan, but regarding the 2d to 3d conversion, if your data set is missing the required overlap, you get get a hole in your output. There is no missing information report during the conversion, nor does it tell you which segments are needed to my knowledge. It is possible to convert large areas in 'Chunks' using PhotoScan, so if you end up with spots that are missing the required information, you can re-sample a smaller area and add the missing photos to your data set, and then reprocess that chunk. The final output is created by aligning the chunks. You can also speed up the process by chunking up the data, which is recommended by Agisoft.
@Darius, correct, to get a 3-d output, you have to overlap enough to get your intended objects on more than one photo. If you are just wanting to align photos, the processing time is much less. I have not used PhotoScan except to generate 3-d output, so I am unsure even if you would need it just to align photos, there are much less expensive solutions I am sure.
@Patrick, thank you for your kind reply.
Pls tell me if an intelligent 2D > 3D conversion algorithm can request missing overlapping parts/ images in interactive mode.
Years ago I was developing 2D car navigation system at Maemo Project by Nokia (based on Debian by late Ian Murdock) since 3D imaginery was available via Lidar scan only.
Audi and others tried to develop 3D car navigation and having joined Google Maps
as independent developer, I was assured to wait for Google Earth based 3D car navigation system.
Since 3D in Google Earth landscape view is as is, I was happy to learn Samsung to develop and offer 2D > 3D live video conversion embedded into LED TV set by default.
To my understanding to get 3D output from ortho 2d images a single building should be
side (at specific angle) scanned to get every wall texture projected and saved.
Downward pointing camera can scan and save 2 wall textures of a single building only, so Sketchup developed a nice way to build 3D architecture environment from 2D projected side wall textures.Missing side wall textures is replaced by some random
milky white texture.
Samsung offers 2D > 3D live video conversion algorithm/s to work fine in landscape view
to extract Z-depth dimension.
I am not sure how 3D buildings, objects get reconstructed from partly only overlapped ortho images.
So, in my opinion, smart 2D > 3D converter should request more ortho images if available, in case of 2D ortho video recording (requesting and self-selecting individual video frames from video container).
Autodesk offers 2D > 3D conversion from 2D video/ images acquired by multi-point mobile cameras.
BTW
Earth's total land area totals < 150 mln km2 so 15,000,000,000 ha
so a lot of scan job still to be done.
Not sure if Google Earth renamed to Alphabet Inc. is interested to provide higher resolution imaginery, overloading its servers and Internet bandwidth in a future
@RG, if you are using Agisoft Photoscan, my experience is that the processing is faster for photo alignment if you have 'Referenced Cameras' and 'Referenced Chunks' with geotagging. The alignment algorithm has to do less work if you have every camera referenced, rather than just having ground markers. After the photos are aligned though, there is no real difference that I have observed in the processing time, and most of the time, probably 90% is post-processing to produce the 3-d images.
@Patrick,
could you explain me how 3-d images are generated from 2D orthos if overlay is limited to nn% (< 50%) ?
100 ha 377 photo with 12 Mpx camera at 150m (500 feet) Overlap and Sidelap 70% GR 5cm/pixel
each photo jpg 5 MB (overage)
1000ha= 3.770 photo 23.000ha= 86.710 photo
87.710 x 5 Mb = 433.550 MB = 433.5 GB
Try to work out 3700 photo in Photoscan and after open the orthophoto, is'n so easy.
My 2 cents.
We are developing a mapping drone in Estonia and flying with it for mapping service.
1) gathering the information is only the first step. We have quite a lot of computing power available, and it still takes us 3 days to get 15 km2 orthomosaic produced.
2) the number of photos will grow very fast when you increase the resolution. So you should aim at the lowest resolution possible.
Some figures on the number of photos are here: http://www.fotoglider.ee/aerial-mapping-service/
We have made some surveys for Estonian Agricultural Registers and Information Board and they required 20 cm/px. We flew at ca 400 m and made 10 cm/px photos for that (to be sure we are at required resolution after image processing).
3) even if you use cloud computing, it not for free. I advise to test out what is the lowest resolution you will need, before you actually start the project - you can save a lot of money on computing on a project of this size.
4) A lot depends on the accuracy of the aerial map needed - to get is "straight" you will need camera with IMU or a lot reference points on the grounds. Most "self made" UAV-s don't carry IMU.
Here is also the main advantage of the manned aircraft - the camera they can carry will give accurate results with just a few reference points on the ground and you will need less computing time.
UAV-s have advantage if yo need to cover small areas. I would think that on this scale manned aircraft is cheaper because of significantly lower image computing cost and less (or non) reference points needed on the ground. As mentioned in this thread: they can cover the area in hours and perhaps you get more comparable results.
-
1
-
2
-
3
-
4
-
5
of 9 Next