UAS Time, WorkflowPosted by Norman_Oklahoma on December 26, 2019 at 4:13 pm
Suppose you have a few hundred photos of a building site (say 100′ square) to process. How long does it take to get from raw photos to orthophoto and 3d point cloud? And how much of that time is your time inputting and identifying GCPs and how much is unattended computer processing time?
- 15 Replies
- MemberDecember 26, 2019 at 8:21 pm
For 100 pics, about 20 mins. Pix4D is what I use and its pretty quick as long as your machine can keep up with it.
We just finished running a large solar farm site, about 18,750 – 24 megapixel photos and about 90 GCPs. That took about 3 days to process the images and another 3 days for the .las file.
- MemberDecember 26, 2019 at 11:34 pm
100 pictures is a pretty small flight, about 10-20 mins to fly. I imagine it would cover 2-3 acres or so. You probably only need 5 targets or so, depending on terrain, etc. If you have the values in a text file using “real” coordinates they will show up in Pix4D close to correct and you can find them with ease. Also, a good trick to to shoot the targets in a methodical order – say in the same direction the images were captured. That will help you find the targets much faster; ie. target 1 will be near pictures 1-15.
But, yes, once you get the hang of the software a 100 image flight should take no more than 1 hour total office time and 30 flight. But bill for at least half days work of standard crew rate. FYI, all of our drone projects bill at a min. of 750.00, even if they just want a picture of of everyone waving for a Christmas card.
- MemberDecember 30, 2019 at 3:15 am
probably add some time ,30-60 minutes depending on the height and spread, for removing artefacts in the orthoimage if there are buildings within your project area.
- MemberDecember 30, 2019 at 7:05 pm
Does the complexity of the thing being modelled affect the processing time?
- MemberDecember 30, 2019 at 7:14 pm
I would not use a drone to extract those features. You can though use a drone for the georeferenced image and them import that into CAD and simply trace for X/Y.
I am sorry already if you need to chase those rebar lines.
I’d bust of the P40 and scan that mama.
- MemberDecember 30, 2019 at 7:43 pm
I have had it done both ways. The scanner is clearly the best way to get a 3d model, but onsite time has been longer than for a drone flight. The concrete is poured on these things within hours of the layout being completed. In fact, the last minute adjustments are ongoing even as the concrete trucks roll up. So we have a very short window of time. Generally we have the deck after 3pm (and it’s dark by 4pm this time of year) with the pour scheduled for the next morning.
The drone is best for an orthophoto, which is generally sufficient. But the processing times have been rather long. Much longer than you fellows have suggested. We want that photo to compare to plan before concrete goes on, so that any necessary adjustments can be made.
- MemberDecember 31, 2019 at 11:28 am
10-4. I understand. Do you use the same targets for each pour? If not I would set some off site targets and never change them. Flight time should be the same 15-20 mins. Then processing shouldn’t be that long. I wonder if the processing time is machine related? Are you pushing data to a cloud or working locally?
- MemberDecember 31, 2019 at 4:07 pm
There is typically only one concrete deck, one pour. There will be 4 floors of wood framing going on top of this deck. I maintain project control offsite, typically – as in this case – across the street. Maintaining photo points anywhere within the site would be a losing proposition. The building footprint occupies 95% of the property area, the sidewalk and adjoining street are a cluster F of laydown materials, forklifts, tradesman’s vehicles, job shack, etc. I set target stickers for control on the deck form panels just prior to flight.
You have given me time estimates for 100 photos. If the overlap was greater and there were 1000 photos to stitch for the same area would the processing time be 10x greater?
- MemberDecember 31, 2019 at 4:30 pm
regarding site control…We perform volume measurements every month for a client and use the same ground control points. They are offsite but always in the same data set and yields reliable results. Not sure your AGL but if you are pushing 250+ you should be able to see targets across the streets. you can always fly a larger area but only process a smaller portion for the final image. Another option would be to use roof corners if you can get a reflectorless shot on them.
100 pics verses 1000 – I have not testing it, but I would assume the processing time would be linear as long as the area is not increasing, which would require more targets.
You can save time but reducing to 75/75 front and side lap. If you are only looking for a pic you don’t need to run double grid, that can also save time. Double grids only help with oblique.
And as a salesman would say…this rub is why they make PPK drones. Keep track of the time spent on GCPs and the repeated headache – upgrading the aircraft to a PPK model may be cost effective on a project like this that has a guaranteed invoice.
A Matrice 210 RTK v2 may be the perfect tool for you. We had a Matrice 210 and it was a bad ass MF. It is a solid unit. We dumped it only because of Federal worries with DJI units being used on Federal installations – which would have had huge impacts for us.
- MemberDecember 31, 2019 at 4:53 pm
There are too many variables to determine processing time.
- Camera type and size of the image
- The processing software
- Computer cpu, speed and core
- Processing needed such as, ortho, LAS, DTM, DSM, contours, etc.
- MemberJanuary 2, 2020 at 9:34 pm
Puget Systems has great articles and benchmarks for Pix4D. Here is their latest article comparing the latest Intel and AMD processors:
- MemberJanuary 6, 2020 at 3:29 pm
We have a decent computer with Pix4D where I process our flights. A 100 picture flight is very small and this ‘beast’ can process it in 15-20 mins. Say we established 5 GCPs throughout and that takes 10 mins to identify and tag. Point clouds and ortho tiles are generated in their respective folders.
- MemberJanuary 6, 2020 at 8:19 pm
I agree with pretty much all of what has been said here. Even with a so-so computer, 100 or so photos won’t take long for the initial processing. Depending on how well your GCP info is organized, in order to bring in them in and correctly locate them, is probably 10 miniatures per GCP, and then a bit to re-optimize. For your project, unless it requires locating in the rel world, I’d probably just use on site scale constraints. That is unless you will be coming back for comparison flights.
As easy as it all is, I’d just figure a couple of hours start to finish with 3/4 of that being processing. 1000 images will not take 10x the time. Maybe double. You won’t have any more GCP’s since it’s the same area, just a lot more tie points for the software to process. This is where a good pc really makes a difference. 800 images on a so-so pc will take several hours, maybe even per step, whereas with a good dedicated pc, it won’t take near that. The nice thing about a good pc, is that you are not tempted to skimp. Crank up the overlap and drop the AGL and get a real good gsd.
Looks like fun though.
Log in to reply.