UAV Photo to 3D accuracy

[JADAstudio]

I'm a week into the UAV Photo to 3D workflow and it really seems like magic. Truly fantastic.

My initial goals are to get orthophotos and point clouds of project sites from 0.25 to 25 acres in size. Residential projects, mostly, but I could see this being used for commercial jobs during conceptual development / site evaluation.

Before I start using this on paying jobs I need to be able to answer the question "Ok, so how accurate is this?". Based on poking around the point cloud and comparing it with 1) other elevation datasets such as the NED and Google Earth, and 2) other point clouds generated from the same photos via competing services It seems the answer is -- at this point, for me -- "not enough to be useful".

So how do I get better at this? What do I need to do to get useful accuracy in a point cloud of a 25 acre site. Useful accuracy for me would be +/- 12 inch in the vertical, preferably better. Will the Photo to 3D service support this kind of accuracy with the appropriate inputs or is that asking to much of photogrammetry in 2017?

Ground control points? Can I rent a handheld GPS on a stick and do my own GCPs, or is that beyond the skillset of your typical nerdy architect?

Better UAV / camera? Currently I'm using Map Pilot on iOS to fly a Phantom 3 Standard which, I know, is the "budget" model. Will a Phantom 4 Pro make that much of a difference?

[vidanom]

Hi @JADAstudio,

Better the camera the more accurate the model will be.

The best results we have seen are those which were taken from a professional camera, which can be a problem for the UAVs to carry.

Some good packaged UAV systems are the Sony camera on the 3DR solo, or the X5S on the Inspire from DJI.

Mitko

[JADAstudio]

A better camera will produce a more detailed model, but won't necessarily improve the large scale (over long distances) accuracy of the point cloud, correct?

[vidanom]

It will be definitely better than using a lower quality camera. 

That's the thing, the more pixels you can get from the details you want to reconstruct the more accurate that 3d model would be. When it comes to a large scale, because of the photo limitation of 250 images, you have to go a little bit higher to get plenty of coverage and overlap and with that you loose some details.

Hopefuly soon we'll increase the limit of 250 images so the quality for these larger scale scenarios can be improved.  

Mitko

[JADAstudio]

Thanks, Mitko. Much appreciated!

A better quality camera is on the shopping list, but I'm not unhappy with the results I'm getting with the P3S, particularly with how inexpensive it was. But, wow, that custom Sony imager on the 3DR Solo looks nice. Too bad it's a $10k+ UAV.

Based on further research it looks like establishing ground control points is my next priority. The challenge is going to be finding something that is "accurate enough" without spending $10,000 on survey equipment. Trimble Catalyst, Emlid Reach, Bad Elf GNSS Surveyor, and RTKLIB (and requisite DIY GNSS data loggers) all look like they have merit. I still have lots to learn.

Thanks for your input!

[Bill.Neuhauser · Accepted solution]

Accuracy first with RECAP 360 can only be obtained with with a RTK or PPK UAV  ie. Topcon eBee or with surveyed in GCPs using a survey grade GPS or Total stations using known BMs.  As for cameras the best inexpensive UAV is the DJI P4 Pro, hands down for $1500 nothing can touch it. Stay far far away from the 3DR with the Sony whatever it is camera for the 10k!  Was outdated a year before it came out! And the camera is I believe a 20mb unit just like the P4P with I believe the same sensor size.   The apps available for the P4P blow away anything out there and I believe from your earlier post are using the best mission app sold  (Map Pilot made by Maps Made Easy.

So as of right now the cheapest answer is the P4P with GCPs.  The expensive answer is the eBee for 20K plus.    Also keep in mind there are other photogrametry software out there that are survey grade and ten times easier to inter your GCP info.   The way Recap 360 sits it is a great TOY, but by no means ready for survey grade work.  The only thing it has going for it is that it is cloud based and processing is really quick compared (once you data is finally uploaded) to whats out there, the down side is that it takes way too long to upload and even longer to input your GCPs. (One head can literately explode during this process).  And there is no report process to use to check its results. And lets not even mention the 250 photo limit.

I like many others has gotten on our soapboxes and expressed our opinions but until someone at Autodesk actually listens and changes Recap it is just a conceptual tool at best.

Bill N

Current part 107 UAV pilot and former 333 exemption holder and full time Professional Engineer.

[JADAstudio]

Thanks, Bill. Good information. A P4P and some kind of GNSS GCP equipment are on the list. Currently leaning toward a pair of Emlid Reach RS units.

And, yes, I seem to find a new UAV-to-mapping service every few days. Lots of good options out there. Do you have a personal preference for orthophoto mapping and/or site-wide point cloud generation? It seems like pricing still varies widely; I'm guessing the industry is still trying to figure out what these kinds of products and services are actually worth.

I can see your points regarding the limitations of ReCap. Hopefully Autodesk is in this segment for the long haul and can devote the necessary time and development resources to make ReCap truly top-tier. Or perhaps they will follow the acquire-and-integrate path as they have done in the past.

Thanks again for your thoughts!

[jimdow528]

Regardless of the GPS/Camera/Ground Control you can improve your quality of results by flying curved flight lines.

BEaWARE of the systematic SfM doming (elevation) error which frequently occurs based on the accumulation of lens calibration errors during the processing of image blocks flown using traditional (linear/parallel) flight lines.  Google "Gently Curving, Convergent, Non-traditional Drone Flight Lines" for a description of curved flight lines.  Justification for the use of curved flight lines can be found in the following research:

Minimising systematic error surfaces in digital elevation models using oblique convergent imagery

Rene Wackrow 

Jim H. Chandler

First published: 16 March 2011

https://doi.org/10.1111/j.1477-9730.2011.00623.x

Results of the simulation process, the laboratory test and the practical test are reported in this paper and demonstrate that an oblique convergent image configuration eradicates the systematic error surfaces which result from inaccurate lens distortion parameters. This approach is significant because by removing the need for an accurate lens model it effectively improves the accuracies of digital surface representations derived using consumer‐grade digital cameras. Carefully selected image configurations could therefore provide new opportunities for improving the quality of photogrammetrically acquired data.

I hope this helps.  Using the attachments, compare traditional (linear/parallel) flight lines with non-traditional (non-linear/non-parallel) flight lines.

Jim Dow

[enssea.algerie]

I like numerous others has gotten on our soapboxes and communicated our sentiments however until the point when somebody at Autodesk really tunes in and changes Recap it is only a theoretical apparatus, best case scenario.