How many GCP's?


Hi,

I am fairly new to the UAV mapping world, I’ve been doing manned aerial surveys for the past 3 years with digital, film, and lidar sensors, though they are similar, there are definitely differences. 

We’ve been mapping fairly square and rectangular parcels with a phantom 3 with varying degrees of success. 

We have a corridor project of a 4 mile long winding road that has a lot of terrain relief. We need to map the road and 25 feet from centerline on both sides.We’re flying this with a phantom 3

How many ground control points and where would be optimal locations to place them? I’ve read the more the better and staggering them is ok but pairing the GCPs in the corridor is best. We are hoping for 2-3 inch accuracy. See attached image for the corridor. Right now I am up to 27 GCPs staggering them on either side of the road and 51 GCP total if I want to have areas that have GCP in pairs. It seems like this could be overkill but I don’t know.

Any help would be much appreciated. Thanks!

Are you the Ensign in Utah by any chance?

I’ve had good results for longer jobs with way less staggered GCP’s, but I’ve never modeled mountains.   

The 3D reconstruction needs precision first (from High Overlap & MTP’s) before you can achieve total accuracy.  The GCP’s only geolocate the model.  

But I’m also a fan of overkill on the field work so you don’t have to go back.  It literally only takes a few seconds per GCP with RTK GPS.     

I’ve flown all of Willard Bay dike three times. 14 miles each way. I staggered the targets every 500’, flew 1000’ each way, there and back with a very high overlap and continued bunny hopping my targets. I had no problem doing it this way. The only down fall is it was very time consuming. I could do around 6.5 miles in a ten hour day.

Thanks for all your replies!

Erik- we are the Ensign in Utah.

Pix4D got back to me this morning with this question and they recommended a minimum of 3 GCPs per flight. I have estimated 9 flights and with 3 GCP per flight, we’d need 27 at minimum. That’s exactly what we initially planned for but it seemed like overkill, but Pix4D confirmed that this is what would be required for this type of flight. 

Thanks!

 

@Ensign Engineering - Just out of curiosity, did you include (2) additional RTK points for every (3) GCP to be used as checks in Pix4D?

And if, yes, how did things turn out?

I am also curious about how things turned out and were additional points used for checks and if so how many?

I have looked through:

https://community.pix4d.com/t/5081-Number-and-distribution-of-GCPs-in-Corridor-Mapping-?page=1#community_comment_115004227606&gsc.tab=0

https://support.pix4d.com/hc/en-us/articles/202559299-Number-and-Distribution-of-Ground-Control-Points-GCPs-in-Corridor-Mapping#gsc.tab=0

https://support.pix4d.com/hc/en-us/articles/202557459#label6&gsc.tab=0

It is hard to determine the amount of GCPs to utilize in corridor mapping.

I am looking at a corridor that is 3 miles (5km) by 200 feet (60 meters).

We are hoping to get +/-0.10 feet (+/- 3 cm) of accuracy.

Hello!

We are planning to survey 300 meters wide to 85 km long corridor. The survey is requested for a detailed design of a future road.

The terrain is mostly flat. We are targeted to GSD of 4cm accuracy to furnish 1:500 scale map as a final deliverable.

We plan to use fixed wing with 85% frontal and 65% side overlap. Each flight will cover 90 Ha (3000m x 300m) and we plan to have 6 GCPs per flight always leaving 2 of the GCPs untouched for the overlapping next flight. My only worry is the number of GCPs? Are 6 GCPs enough?

I suggest you consider using a ground based survey method for the detailed road design (terrestrial 3D scanner, RTK GPS, etc) 

You will need MANY GCP’s and check points to verify the Contour data produced is accurate.  

The amount of Man Hours for the Geometric and Earthwork Design outweigh the survey effort in orders of magnitude. 

Small errors in your contour data become very expensive problems during construction.   

When your deliverable will directly impact an Engineering Design, you must validate your results with “ground truth”. 

QA will require almost as much time as a Strip Topo w/ RTK GPS would.   

In my opinion, an aerial survey can produce a very useful Ortho to Compliment your ground survey in a Road Design Project.  

 

 

@ Ryan Fontaine,

What is in your opinion the main reason for the Photogrammetry Survey to fail producing an accurate DSM and DTM?

The terrain is mostly flat. I plan to have 85% to 65% frontal and side overlap respectively at 100 m altitude. Calculated GSD is below 4cm. 6 GCPs per 3000 m length using 2 same GCPs for adjacent sections plus dozens of MTPs to insure good reproduction.

Pix4D can obviously produce an accurate reconstruction and it’s a wonderful tool.  My comments were geared toward your deliverable being used for Final Roadway Design.   

 

 

@ Ryan Fontaine,

Thank you for sharing your opinion.

I actually counted on Pix4D to perform the project with limited budget and time.

However you suggest that Photogrammetry Technology is not yet capable of producing satisfactory deliverable for Detailed Road Design without spending as much time as a Strip Topo with RTK GPS, which is frustrating.

Is LIDAR the only option to have the “ground truth”?

Per this article : https://support.pix4d.com/hc/en-us/articles/202558889-Accuracy-of-Pix4Dmapper-Outputs#gsc.tab=0 

Generally, the expected accuracy of a correctly reconstructed project is:

  • 1-2 GSD horizontally (X,Y coordinates). 
  • 1-3 GSD vertically (Z coordinate).

Your planned 4cm GSD = 4cm to 12cm Vertical Accuracy (10 to 30 inches).

Does 12 cm Accuracy fit within your Scope of Work ?  If it doesn’t, then you need to decrease your flight altitude.  

If you decide to fly this project, you can consider using Pix4dTagger for automatic target recognition.  It saves a lot of work.  Add many extra targets to be used for Manual Check Points ( essentially a GCP that’s not used in the model reconstruction).  If you can accept the implied accuracy after comparing the Manual Check Point’s computed positions, re-run the model while using the Manual Check Point Targets as additional GCP’s.  

 

Even a LiDaR survey requires verification.  

1 Like