3DR claim of <1cm accuracy with sitescan witho̶u̶t̶ RTK

Trying to enhance my photogrammetry accuracy with Reach being at the top of the list to get me there. Also keeping a keen eye on turnkey solutions provided by the likes of 3DR with their Sitescan (solo and sitescan app and pix4d engine) and trying to see the value of a $8000 yearly license.

Well yesterday I received a white paper case study reporting an accuracy of <1cm both vertical and horizontal. Which I find totally implausible. not even with RTK which there is no mention of.


Do these guys need to be called out? They’re heavily within the construction industry with turnkey solutions aidedby autodesk and BIM integration which is probably where their fees can be accounted for. I think the whole turnkey solution to be largely misguided but it’s an interesting space.
Here’s a link to that white paper.


So only 2 check points for the site? Note that the control points used to reference the model are used to quote residuals. What about points between the control points? This is where the error will lie.

To quote a mean RMSE is meaningless. You could be -5m to +5m error and your average would be 0m!

Thanks for cutting through the jargon. Yes totally useless and utterly misleading figure. Misleading the industry and depriving the professionals. The industry relies on safety and accountability trails. The whole point of these turnkey systems to the industry is to provide a rigorous accuracy reports. Anyway I can’t see them lasting long along this route.

Actually the UAS and software setup used in $12,000. A solo with the nice new Sony R10C. Very expensive. Again it’s all about the software integration with current major players like autodesk BIM. But that integration is very simple and actually not that great.

Pix4d also offer a platform Pix4DBIM which incorporates a timeline and project management system independent of autodesk BIM which runs at EUR4000/yr which would seem the more sensible route to an independent mapping company.

I’m also looking into opensource solutions involving VisualSFM and Mesher. To completely remove software overheads. Clients have their own project management solutions so just delivering data is all I require.

I’d be interested to hear what route other professionals are taking.

1 Like

Having read the white paper it is very thin on detail.

No information on the survey methods used to set the GCP.

No details of the mean error of the GPS and Check points collected on site

No details on the calibration of the camera

It makes me uneasy when a highly skilled job can be boiled down by the marketing men into;

  1. Go to site with drone
  2. Collect accurate data
  3. Process using our fancy software
  4. Cash huge cheque from client

Ask any surveyor what is the difference between accuracy and precision?

accuracy of <1cm to what? In surveying accuracy is a relative term.

End of rant!

1 Like

Reminds me of a old Russian joke I learned. It was originally about the so-called “New Rich” in Russia after the collapse of the USSR. It seems quite timely here. This is a revised version
Two UAV surveyors meet on on a construction site, each carrying their surveying equipment, drones, and cameras. Bob is carrying two Emlid Reach RS’s, a Solo drone, and a smartphone. All his software is opensource, QGIS, Cloud Compare, Visual SFM, etc. All these produce standard models, point clouds, rasters, etc. that can with one click be imported into the software of your choice, including all Autodesk design programs.
Terry, on the other hand has the same Solo Drone (only more stickers), a tablet with a fancy hood, and most importantly, all top dollar processing software that apparently seamlessly integrates with Autodesk software.(Terry’s company uses ARCgis)
Terry looks at Bob and says inquisitively. "What’s that nice looking equipment you’re carrying?"
to which Bob replies. “My UAV and RTK survey equipment for setting ground control points?”
“How much did it cost you?” responds Terry nervously.
“Let’s see, all together around 3000 bucks when you include the tripod and the poles.” Bob answers, adding, “all my software is open source so I have no software expenses on top of this.”
“You fool!” laughs Terry derisively, rubbing his hands together in glee. "I hire a surveyor for ground control points and for the UAV I got the same setup as you and it does the exact same thing and I pay $12000 annually!

I remember trying to understand this joke when we first moved to Russia, and thought it was some quaint nuance of the “Russian soul” :slight_smile: but it seems to apply quite well in today’s construction industry as well. Take it for what it’s worth. Hopefully a little joke doesn’t break forum rules.
I am interested in Opensource solutions as well.
There are opensource options and then there are options that offer free services for small surveys.
Low cost and free.
1.I have used mapsmadeeasy.com for this. They offer free image processing for small flights and after that it’s a pay per use so it’s not a big investment to get started. The also have a flight planner for iOS that has given me the best results with my DJI mavic. One advantage with them is that you can use ground control points without stepping into a high dollar software. Also, you can download point clouds, etc.
2. Free-I recently tried CloudCompare for working with point clouds. I was really impressed.
3. Lo-Cost Then there’s Autodesk’s own Recap, which used to be in site scan but I’m not sure anymore. This will process the imagery from your camera and can be used for georeferencing point clouds.Site Scan has been advertising more about an integration with pix4d. I’m not sure about the connection.
4. Free-Drone Deploy’s free version(expired trial). My experience has been good with drone deploy but they really limit what you can download in the free version. No cheap solution for ground control points.
5. Free- QGIS can generate contour lines from DEMS (from image processing software). QGIS is a great place to analyze, and convert much more GIS information. shp to dxf, convert coordinate systems, etc.Also hydrology with GRASS plugins, etc.

Okay, I’ve wandered a bit from the topic of the thread but I’d sure be interested in feedback on this and other software.P.S. I work with Autodesk software, Revit, Infraworks, and sometimes Civil 3D a little and I don’t see a benefit in paying big bucks to “integrate” them with your Site Scan software, but in all fairness, I must add that I’ve never seen Site Scan in use, either, other than a few youtube videos.


You missed a few:
OpenDroneMap for the SFM processing, opensource. with its excellent add on WebODM
3DF Zephyr free which is closed source but free for up to 50 images.
Colmap, opensource SFM, which takes you through to a textured mesh using CUDA accelleration.
Regard 3D opensource SFM

Point and mesh manipulation

Back on topic.
A few years ago I was working with people doing underwater surveys using video for odometry along the seabed.
So with a very well calibrated camera and no dynamic camera calibration it is possible to use the features seen between images to determine movement. This is the basis of structure from motion, So if you can ensure your processing method introduces no distortion then you can scale the entire point cloud created to fit between two points and make measurements on the entire point cloud. The key is ‘no distortion’ which is down to camera calibration and a fairly rigid survey method that introduces enough contrasting views to create an even ‘tension’ in the final result.

1 Like

Having now read the paper properly. Your title for the post is in error as they clearly used RTK for the GCP’s
I have no issues with their findings.
Having nine GCPs to cm level (or better) accuracy within a flat survey area of the size shown will with care deliver the accuracies stated without RTK on the drone as long as you use a good fixed camera calibration and fly a good grid.

1 Like

Right yes GCP’s are there. No mention of RTK directly, my mistake. Didn’t realise the solo has a RTK receiver. Bit more work involved than suggested I’m guessing though. How long would you expect to have to wait for RTK “calibration”?

The Solo doesn’t have an RTK receiver.
A highly underpaid and undervalued, possibly even unloved, individual (surveyor) goes around and ‘fixes’ the positions of the Ground Control Points using an RTK on a pole (like Reach RS) these points which have high integrity and accuracy are used within the drone image processing to tie the subsequent surfaces and images to the real world. It is a standard practise. With RTK on the drone you can reduce the need for ground control points which are time consuming to put in. IN some cases you can do away with them completely. (Flame me here) if you are mapping a remote area or swamp that does not allow the installation of permanent marks.

I use a large target launch pad and wait at least 5 mins before launching with the drone sitting in the very middle of the target. That gives me a single GCP, to confirm the elevation and absolute position (of that point). As a minimum I will add one more GCP ( at a good distance from the first) to confirm scale and slope.

The real challenges come into play when you start really caring about the third dimension.i.e. 3D structures.


Thanks Simon, clearing things up for me a little. There is an aftermarket RTK receiver available for the solo http://diydrones.com/m/blogpost?id=705844%3ABlogPost%3A2344182 but right I doubt if that’s part of the site scan kit.
I hadn’t realised taking GCP’s made a rover RTK receiver unnecessary to achieve high accuracy. So then your GCP receiver is then set as a base unit with another rover calculating corrections or a rover utilizing corrections from an NTRIP network?
Curious to know how utilizing a NTRIP network for a UAV RTK solution can streamline things.
I also wonder if the solo used in this is providing gimbal/camera attitude via gimbal encoders/IMU as I heard Mr Emlid himself describe this being used to achieve accuracy in photogrammetry software. And if this is being used how are DIY folk getting this info. Uart serial_api gimbal control via alexmos?
and one more nagging question Simon those GCP’s you take, do you have a set displacement you apply for the precise location of the receiver on the drone?

Two thumbs up for providing an opportunity to read English translated and geodesy-adopted version of the anecdote about “new rich Russians” :smile:

1 Like

RTK on the drone reduces the need for GCPs
If you look at the 3dr paper 9 GCP! They must be 150m apart. as said in post I can live with two, one of which is my drone launch point, or only one (drone launch point ) for difficult to access sites.

You don’t need the gimbal orientation as the image is ‘fitted’ by the features identified and the camera position. What is good practise is to put a bit of ‘tension’ between the images by pointing the camer 10 degrees off vertical and doing some cross lines. Oh and NEVER use auto camera calibration.

Really quite impressed with mapsmadeeasy.com, Love the business model, low entry point, per job scalable.
Here’s a set of image I took a while back which aren’t very good about as bad a dataset you could ask for. hackedup chinese camera on a flamewheel navio UAV. focus over field a bit patchy, no trigger sync, images geo- referenced loosely (camera on timer) but the results are fairly skookum. Much better than what I can get out of VisualSFM.
Browser 3d model https://skfb.ly/6uJoJ

1 Like

Altizure have a slightly different model.
You can upload and view online a model of almost any size, but if you want the results then you pay pro-rata per gigapixel. Combined with WebODM this gives me the option to create products for use elsewhere, but display in a great 3D model.

Altizure do a better job of defining structures and man made features as their focus is on 3D, they do not just rely on a Poisson mesh like LLE.

I’m looking for something that can create package a model, simple to use navigation with switchable layers, measuring tools, and ability to to add notes and clickable elements that bring up info boxes. WebODM sounds perfect. Trying out altizure with a decent set of images but I don’t think it’s liking the size of my images at 36mp 7mb jpegs, stuck at processing images stage.

Did altizure get there?
The Potree cloud viewer will do what you want

Yeah in the end it got there. I reduced the pics to 12mp as it was quite costly for this job. Here’s the results. https://www.altizure.com/project/5a79d75dd7a1f3053a2177ae/model
And compared to dronedeploy

Altizure definitely takes it quite substantially. I didn’t bother with pix4d cloud as it’s seemed to be the worst in previous tests. Although I suspect desktop version will take much better results.

Wow, yes Altizure does nail it in comparison.
I like Altizure when the client doesn’t mind public delivery as it is a great showcase.

Yeah I like the vibe and the place they’re coming from. They’re a bunch of Hong Kong Uni professors and students who aren’t salesmen but have a marketing strategy that should work and allow them to sharpen. I like the price per job aspect also. Although I’m a little trepidatous to upload full Res files (and they’re not exactly perfectly captured tak sharp so…) at three times the cost, but feel I asked to test they’d let me. Their support is also really quick and seemingly 24hr. Great recommendation Simon thanks.

This topic was automatically closed 100 days after the last reply. New replies are no longer allowed.